From 9ced91e6d7a3daa820ba20aed8d6f8ab68228bb0 Mon Sep 17 00:00:00 2001
From: abhina-kumar <151630489+abhina-kumar@users.noreply.github.com>
Date: Mon, 1 Sep 2025 00:39:05 +0530
Subject: [PATCH 01/13] Add explainer for Semantic Web Tools using Well-known
 Destinations

---
 ...web-tools-using-well-known-destinations.md | 357 ++++++++++++++++++
 1 file changed, 357 insertions(+)
 create mode 100644 explainers/semantic-web-tools-using-well-known-destinations.md

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
new file mode 100644
index 0000000..5a66a2c
--- /dev/null
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -0,0 +1,357 @@
+# WAI-Adapt: Semantic Web Tools for Agentic AI Using Well-known Destinations Explainer
+
+## Authors
+
+- Abhinav Kumar (@abhina-kumar), SAP Labs India Pvt. Ltd.
+- Matthew Tylee Atkinson (@matatk), Samsung Electronics
+- Lionel Wolberger (@lwolberg), Level Access
+- Janina Sajka (@JaninaSajka), W3C Invited Experts
+
+## Participate
+
+* Issues: https://github.com/w3c/adapt/issues
+* Discussions: https://github.com/w3c/adapt/discussions
+
+## Contents
+
+<!-- toc -->
+
+- [Introduction](#introduction)
+- [Current State: The API Dependency Challenge](#current-state-the-api-dependency-challenge)
+- [The Semantic Web Tools Solution](#the-semantic-web-tools-solution)
+- [Scope and Limitations](#scope-and-limitations)
+- [Real-World Applications](#real-world-applications)
+- [Building on Well-known Destinations](#building-on-well-known-destinations)
+- [How It Works: The Technical Foundation](#how-it-works-the-technical-foundation)
+- [Semantic Web Tools Architecture](#semantic-web-tools-architecture)
+  * [Semantic Web Tools Capabilities](#semantic-web-tools-capabilities)
+  * [Discovery and Navigation Mechanisms](#discovery-and-navigation-mechanisms)
+  * [MCP Server Integration](#mcp-server-integration)
+- [Agentic AI Integration Patterns](#agentic-ai-integration-patterns)
+- [Division of Responsibilities: LLM vs Semantic Web Tools](#division-of-responsibilities-llm-vs-semantic-web-tools)
+- [Extended Destination Types for AI Tools](#extended-destination-types-for-ai-tools)
+- [Open Questions](#open-questions)
+- [References](#references)
+
+<!-- tocstop -->
+
+## Introduction
+
+**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps.
+
+However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs.
+
+**The Problem Today:**
+When a user asks an AI agent to "find contact information across my service providers," the agent faces a dilemma. For the 20-30% of sites with APIs, it can work reliably. For the remaining 70-80%, it must resort to fragile CSS selectors and HTML parsing that requires constant maintenance.
+
+**Our Proposal:**
+This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) standard to bridge this gap through **Semantic Web Tools**—standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
+
+**The Vision:**
+Instead of agents saying "find the element with class '.contact-info'", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating a new middle ground between full API development and unreliable scraping.
+
+## Current State: The API Dependency Challenge
+
+Today's agentic AI systems face a stark reality: they live in a two-tier world of website interaction.
+
+**The Privileged Few (20-30% of sites):**
+Sites like Twitter, GitHub provide robust APIs. Here, AI agents work beautifully—they can reliably access data, perform actions, and integrate seamlessly into workflows. These sites offer the structured, programmatic access that AI systems need.
+
+**The Struggling Majority (70-80% of sites):**
+Most websites—from local businesses to government agencies—don't provide APIs for basic functions like finding contact information, accessing help resources, or managing user accounts. This forces AI agents into a world of fragile workarounds.
+
+### The Brittle Fallback Reality
+
+When APIs aren't available, AI agents become digital archaeologists, excavating information through:
+
+- **CSS Selector Archaeology**: Hunting for `.contact-info` classes that vanish with the next design update
+- **HTML Structure Divination**: Parsing unstructured content that changes without notice  
+- **Site-Specific Scripting**: Maintaining individual automation for each website—a maintenance nightmare
+
+### Real-World Pain Points
+
+Consider an AI assistant tasked with "auditing accessibility statements across our company's website portfolio." For sites with APIs, this is straightforward. For the majority without APIs, the agent must:
+
+1. **Guess** where accessibility statements might be located
+2. **Hope** the HTML structure hasn't changed since the last update
+3. **Fail gracefully** when selectors break, leaving users with incomplete results
+
+**The Core Problem**: We need a semantic bridge between the reliability of APIs and the universality of web interfaces.
+
+| Approach | Coverage | Reliability | Maintenance Reality |
+|----------|----------|-------------|---------------------|
+| **APIs Only** | ~20-30% of sites | High | Low (when available) |
+| **Traditional Scraping** | ~100% of sites | Very Low | Constant firefighting |
+| **Semantic Web Tools** | Growing with adoption | High | Set-and-forget |
+
+## The Semantic Web Tools Solution
+
+Well-known Destinations offer an elegant solution to this challenge. Instead of websites needing to build full APIs, they can provide semantic signposts that AI agents can follow reliably.
+
+**How It Works:**
+Rather than an AI agent hunting for contact information using fragile selectors like `document.querySelector('.contact-page-link')`, it simply looks for the semantic marker: `rel="contact"`. This approach transforms brittle technical navigation into reliable semantic discovery.
+
+## Scope and Limitations
+
+### What Well-known Destinations Solve Well
+
+* **Content Discovery**: Finding specific types of pages (help, contact, accessibility statements, product catalogs etc.)
+* **Information Extraction**: Extracting data from destination pages
+* **Consistent Navigation**: Standardized navigation patterns across different websites
+
+### What Requires Additional Solutions
+
+* **Complex Actions**: Multi-step workflows like password changes, account modifications, or transaction processing
+* **Authenticated Operations**: Actions requiring user-specific authentication and authorization
+* **Site-specific Workflows**: Unique business processes that don't map to standard destinations
+
+### Human-in-the-Loop and Layered Approaches for Complex Scenarios
+
+For complex scenarios like account management, a layered approach works best:
+
+1. **Well-known Destinations**: Provide navigation to account management areas
+2. **API Integration**: Use APIs where available for sensitive operations
+3. **Human-in-the-Loop**: Include human oversight and intervention for sensitive or complex tasks
+
+The primary approach for sensitive operations should be human-in-the-loop, where AI agents:
+- Navigate to the relevant pages using semantic destinations
+- Extract and present available options to users
+- Hand control to humans for actual execution of sensitive tasks
+- Provide guidance and context to facilitate user completion
+
+## Real-World Applications
+
+Imagine the possibilities when AI agents can reliably navigate any website using semantic markers:
+
+**The Accessibility Auditor**: A compliance officer asks their AI assistant to "check accessibility statements across our 50 partner websites." Instead of manually visiting each site or maintaining fragile scrapers, the AI agent systematically discovers and analyzes `accessibility-statement` destinations across all sites in minutes.
+
+**The Support Research Assistant**: A customer facing issues with multiple service providers asks for help options. The AI agent discovers `contact` and `help` destinations across different platforms, presenting a comprehensive support landscape without breaking when sites redesign.
+
+**The Digital Estate Manager**: A user wants to update their profile information across platforms. The AI agent navigates to `account-settings` destinations on each site, then guides the user through the updates with full context and control.
+
+**The Compliance Monitor**: A legal team needs regular checks of privacy policies across their digital ecosystem. Semantic destinations enable automated discovery and monitoring without the maintenance burden of traditional scraping.
+
+These scenarios share a common pattern: **AI agents handle the discovery and navigation complexity, while humans maintain control over sensitive decisions and actions.**
+
+## Building on Well-known Destinations
+
+The beauty of this approach lies in its foundation: the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need.
+
+**From Human Accessibility to AI Navigation**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents—what helps humans navigate also helps machines understand.
+
+**The Technical Foundation**: Websites already mark important destinations using standard HTML:
+
+```html
+<link rel="contact" href="/contact-us">
+<link rel="help" href="/support">  
+<link rel="accessibility-statement" href="/accessibility">
+```
+
+**Semantic Discovery**: Instead of guessing where functionality might be, AI agents can programmatically discover what each website offers through these semantic markers. This transforms unpredictable navigation into reliable discovery.
+
+**Progressive Enhancement**: Sites don't need to rebuild anything—they simply add semantic markers to their existing pages. This creates a path for gradual ecosystem adoption without disrupting current workflows.
+
+## How It Works: The Technical Foundation
+
+The architecture is elegantly simple, built on three core capabilities:
+
+**1. Destination Discovery**: AI agents enumerate available destinations by reading the semantic markers websites provide—either through HTML `<link>` elements or centralized linkset documents.
+
+**2. Reliable Navigation**: Using discovered destinations, agents convert semantic names like `contact` into actual URLs and navigate confidently to the right pages.
+
+**3. Intelligent Content Processing**: Once at the destination, agents retrieve page content and hand it to Large Language Models (LLMs) for intelligent extraction and analysis.
+
+This division creates a **separation of concerns**: simple tools handle the mechanical navigation, while sophisticated AI handles the intelligent understanding of content.
+
+### The AI Integration Story
+
+The magic happens in how Semantic Web Tools integrate with AI systems through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification)—an open standard that allows AI applications to securely connect with external tools.
+
+**The Workflow**: When a user asks an AI agent to "find support options across my service providers," here's what happens:
+
+1. **Discovery Phase**: The agent uses Semantic Web Tools to discover what destinations each website offers
+2. **Planning Phase**: Based on available destinations, the agent plans the optimal navigation strategy  
+3. **Execution Phase**: The agent navigates to relevant destinations and retrieves content
+4. **Analysis Phase**: AI processes the retrieved content to extract and synthesize useful information
+
+This pattern works consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction.
+
+## Semantic Web Tools Architecture
+
+### Semantic Web Tools Capabilities
+
+Semantic Web Tools provide standardized capabilities that work uniformly across any website implementing Well-known Destinations:
+
+#### Core Capabilities
+
+* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site through parsing `<link>` elements or accessing linkset documents
+* **Semantic Navigation**: Navigating to specific destinations using semantic identifiers rather than site-specific selectors
+* **Content Retrieval**: Fetching page content from destination URLs and delivering it to LLMs for intelligent processing
+
+**Benefits of Semantic Approach:**
+- Single set of capabilities works across all compliant websites
+- No site-specific customization or maintenance is required
+- Consistent, predictable behavior for AI agents
+- Leverages semantic web standards for reliability
+
+**Note**: Content extraction and analysis is handled by LLMs, while form interactions typically require human-in-the-loop approaches for security and usability reasons.
+
+### Discovery and Navigation Mechanisms
+
+Semantic Web Tools leverage the discovery mechanisms already defined in the Well-known Destinations proposal:
+
+#### HTML Link Element Parsing (Well-known Destinations Proposal)
+
+The Well-known Destinations proposal specifies that websites mark destinations using `<link>` elements in the HTML head. Semantic Web Tools utilize this existing mechanism:
+
+```html
+<head>
+  <link rel="accessibility-statement" href="/accessibility">
+  <link rel="help" href="/support">
+  <link rel="log-in" href="/sign-in">
+  <link rel="products" href="/products">
+</head>
+```
+
+#### Centralized Linkset Discovery (Well-known Destinations Alternative)
+
+As an alternative approach explored in the Well-known Destinations proposal, sites implementing RFC 9264 Linksets can provide centralized destination discovery. Semantic Web Tools can leverage this approach when available:
+
+```json
+{
+  "linkset": [
+    {
+      "anchor": "https://example.com/",
+      "accessibility-statement": [{"href": "https://example.com/accessibility"}],
+      "help": [{"href": "https://example.com/support"}],
+      "log-in": [{"href": "https://example.com/sign-in"}]
+    }
+  ]
+}
+```
+
+**Benefits of the Linkset Approach:**
+- Single request to discover all destinations
+- Centralized management of destination mappings
+- Support for complex site hierarchies
+
+#### Future Extensions
+
+While building on the Well-known Destinations foundation, future extensions could define additional discovery mechanisms specifically optimized for automated tools:
+
+```
+GET /.well-known/destinations
+```
+
+This would return a standardized JSON format describing available destinations and any additional metadata needed for automated processing.
+
+### MCP Server Integration
+
+The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification) is an open standard that enables AI applications to securely connect to external systems and data sources. MCP defines a standardized way for AI applications to interact with tools, resources, and services through a simple protocol.
+
+#### MCP Architecture Overview
+
+MCP operates on a client-server model:
+- **MCP Clients**: AI applications (like Claude, ChatGPT, or custom AI agents)
+- **MCP Servers**: Applications that expose tools and resources to AI systems
+- **Protocol**: Standardized JSON-RPC communication between clients and servers
+
+#### Semantic Web Tools as MCP Tools
+
+Semantic Web Tools can be exposed through MCP servers, providing AI agents with standardized access to website interaction capabilities. This integration allows AI systems to discover destinations, navigate to semantic endpoints, and retrieve content for LLM processing through the established MCP protocol.
+
+#### Benefits of MCP Integration
+
+* **Standardized Interface**: AI agents can interact with Semantic Web Tools using standard MCP protocols
+* **Ecosystem**: Leverage the growing MCP ecosystem of tools and integrations
+
+## Agentic AI Integration Patterns
+
+Agentic AI systems integrate with Semantic Web Tools through a straightforward workflow that leverages semantic destinations for reliable website interaction. The integration follows a consistent pattern regardless of the specific task or websites involved.
+
+### Core Integration Workflow
+
+#### 1. Destination Discovery
+AI agents use the Semantic Web Tool to discover available destinations on target websites or pages by parsing well-known destination markers (`<link>` elements, linksets, or well-known URI endpoints).
+
+#### 2. Destination Selection and Planning  
+Based on discovered destinations and user requirements, AI agents plan which destinations to visit and in what sequence to accomplish their tasks.
+
+#### 3. Semantic Navigation and Content Retrieval
+AI agents direct the Semantic Web Tool to navigate to selected destinations and retrieve page content for analysis.
+
+#### 4. LLM Processing and Synthesis
+AI agents process the retrieved content using LLMs to extract relevant information, make decisions, and synthesize results across multiple sites.
+
+### Workflow Characteristics
+
+**Uniform Approach**: The same Semantic Web Tool and workflow pattern works across all websites that implement Well-known Destinations, eliminating the need for site-specific customization.
+
+**Semantic Reliability**: Navigation based on semantic destinations (`contact`, `help`, `accessibility-statement`) rather than fragile selectors ensures consistent operation even when sites update their designs.
+
+**Clear Separation**: Semantic Web Tools handle technical navigation and content retrieval, while LLMs handle intelligent processing and decision-making.
+
+**Scalable Coordination**: AI agents can easily coordinate multi-site workflows by applying the same pattern across multiple websites simultaneously.
+
+This integration pattern enables AI agents to work reliably across any collection of websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
+
+## Division of Responsibilities: LLM vs Semantic Web Tools
+
+The system operates through a clear separation of concerns between the LLM (reasoning/coordination) and Semantic Web Tools (technical execution):
+
+### LLM Responsibilities
+* **Task Planning**: Interpreting user requests and determining which websites and destinations are needed
+* **Workflow Coordination**: Orchestrating the sequence of operations across multiple sites
+* **Cross-site Analysis**: Synthesizing information from multiple sources into coherent responses
+* **Error Handling**: Deciding how to respond when tools encounter failures or missing content
+* **User Interaction**: Managing clarifications, progress updates, and result presentation
+
+### Semantic Web Tools Responsibilities  
+* **Destination Discovery**: Finding available Well-known Destinations on target websites
+* **Navigation**: Converting semantic destination names to URLs and performing HTTP requests
+* **Content Retrieval**: Fetching raw page content and passing it to LLMs for processing
+* **Error Reporting**: Detecting and reporting technical failures (network errors, missing pages, authentication failures) to the LLM for decision-making
+
+### Content Processing Strategy
+
+Semantic Web Tools handle only the technical aspects of content retrieval, while LLMs handle all content processing:
+
+- **Semantic Web Tools**: Navigate to destination URLs and fetch raw HTML content
+- **LLMs**: Extract, clean, structure, and interpret the content based on the specific task requirements
+
+This division ensures that the intelligent processing of unstructured web content is handled by LLMs, which can adapt to different page layouts, content types, and extraction requirements. Protocol issues like rate limiting, authentication failures, or access restrictions are reported to the LLM, which decides how to respond (retry, use alternative destinations, request human intervention, etc.).
+
+## Extended Destination Types for AI Tools
+
+The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating where APIs, schemas, and webhooks don't exist, a few additional destination types could be beneficial. These are examples of potential extensions, as more use cases may be discovered in the future.
+
+### Example Additional Destinations
+
+* **`search`**: Site-specific search functionality when no search API is available
+* **`account-settings`**: User profile/account management pages
+
+These destination types address common agentic AI use cases that cannot be handled through existing Well-known Destinations and where sites lack APIs. Importantly, these additional destinations would also benefit human users by providing clearer navigation paths to search functionality and account management.
+
+As the ecosystem evolves and more agentic AI use cases emerge, additional destination types may be identified and proposed through the standard development process.
+
+## Open Questions
+
+The following questions remain open for extending Well-known Destinations to support agentic AI systems:
+
+### Well-known Destinations Extensions
+
+* **New Destination Types**: Should additional destinations like `search` and `account-settings` be standardized?
+* **Discovery Optimization**: Should Well-known Destinations include a centralized discovery endpoint (e.g., `/.well-known/destinations`) to improve efficiency instead of requiring parsing of individual pages?
+
+### Standardization Process
+
+* **Governance Model**: How should new AI-relevant destination types be proposed, reviewed, and added to the standard?
+
+## References
+
+### Foundational Work
+
+* [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
+* [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
+* [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
+* [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)

From 6805e86450236e5b8ce93bb6a56382defc961bf7 Mon Sep 17 00:00:00 2001
From: abhina-kumar <151630489+abhina-kumar@users.noreply.github.com>
Date: Mon, 1 Sep 2025 01:12:04 +0530
Subject: [PATCH 02/13] Minor Update on explainer

---
 .../semantic-web-tools-using-well-known-destinations.md     | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index 5a66a2c..abd32a2 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -14,8 +14,6 @@
 
 ## Contents
 
-<!-- toc -->
-
 - [Introduction](#introduction)
 - [Current State: The API Dependency Challenge](#current-state-the-api-dependency-challenge)
 - [The Semantic Web Tools Solution](#the-semantic-web-tools-solution)
@@ -33,8 +31,6 @@
 - [Open Questions](#open-questions)
 - [References](#references)
 
-<!-- tocstop -->
-
 ## Introduction
 
 **Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps.
@@ -48,7 +44,7 @@ When a user asks an AI agent to "find contact information across my service prov
 This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) standard to bridge this gap through **Semantic Web Tools**—standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
 
 **The Vision:**
-Instead of agents saying "find the element with class '.contact-info'", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating a new middle ground between full API development and unreliable scraping.
+Instead of agents saying "find the element with class `.contact-info`", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating a new middle ground between full API development and unreliable scraping.
 
 ## Current State: The API Dependency Challenge
 

From 914c6ef45d480435d06b0fa22d71528b28294eff Mon Sep 17 00:00:00 2001
From: abhina-kumar <151630489+abhina-kumar@users.noreply.github.com>
Date: Tue, 2 Sep 2025 20:29:19 +0530
Subject: [PATCH 03/13] Update
 explainers/semantic-web-tools-using-well-known-destinations.md

Co-authored-by: Matthew Tylee Atkinson <matatk@users.noreply.github.com>
---
 explainers/semantic-web-tools-using-well-known-destinations.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index abd32a2..3151a65 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -41,7 +41,7 @@ However, there's a fundamental challenge: while these AI agents excel at working
 When a user asks an AI agent to "find contact information across my service providers," the agent faces a dilemma. For the 20-30% of sites with APIs, it can work reliably. For the remaining 70-80%, it must resort to fragile CSS selectors and HTML parsing that requires constant maintenance.
 
 **Our Proposal:**
-This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) standard to bridge this gap through **Semantic Web Tools**—standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
+This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools**—standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
 
 **The Vision:**
 Instead of agents saying "find the element with class `.contact-info`", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating a new middle ground between full API development and unreliable scraping.

From 0c40d92f583500d1d4ce12c3cb5c05a5ed62bd58 Mon Sep 17 00:00:00 2001
From: abhina-kumar <151630489+abhina-kumar@users.noreply.github.com>
Date: Wed, 1 Oct 2025 11:37:32 +0530
Subject: [PATCH 04/13] Update
 explainers/semantic-web-tools-using-well-known-destinations.md

Co-authored-by: Matthew Tylee Atkinson <matatk@users.noreply.github.com>
---
 explainers/semantic-web-tools-using-well-known-destinations.md | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index 3151a65..bf5afcd 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -3,9 +3,6 @@
 ## Authors
 
 - Abhinav Kumar (@abhina-kumar), SAP Labs India Pvt. Ltd.
-- Matthew Tylee Atkinson (@matatk), Samsung Electronics
-- Lionel Wolberger (@lwolberg), Level Access
-- Janina Sajka (@JaninaSajka), W3C Invited Experts
 
 ## Participate
 

From fa61f821afa9815efb48f000729a2c1244613a50 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Tue, 21 Oct 2025 18:42:36 +0530
Subject: [PATCH 05/13] Enhance introduction and problem statement in Semantic
 Web Tools explainer for improved clarity and accessibility focus

---
 ...web-tools-using-well-known-destinations.md | 118 ++++++++++++------
 1 file changed, 83 insertions(+), 35 deletions(-)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index bf5afcd..e7f0bd6 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -30,18 +30,19 @@
 
 ## Introduction
 
-**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps.
+**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps. For users with disabilities, these AI agents can serve as powerful assistive technologies, helping navigate digital environments and complete tasks that might otherwise be challenging.
 
-However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs.
+However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs, particularly problematic for users who depend on consistent, reliable access to digital services.
 
-**The Problem Today:**
-When a user asks an AI agent to "find contact information across my service providers," the agent faces a dilemma. For the 20-30% of sites with APIs, it can work reliably. For the remaining 70-80%, it must resort to fragile CSS selectors and HTML parsing that requires constant maintenance.
+### The Problem Today
 
-**Our Proposal:**
-This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools**—standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
+When a user with cognitive disabilities asks an AI agent to "find contact information across my service providers," or when someone with motor impairments needs assistance navigating multiple websites, the agent faces a dilemma. For the 20-30% of sites with APIs, it can work reliably. For the remaining 70-80%, it must resort to fragile CSS selectors and HTML parsing that requires constant maintenance, causing unreliable experiences precisely when consistency matters most.
 
-**The Vision:**
-Instead of agents saying "find the element with class `.contact-info`", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating a new middle ground between full API development and unreliable scraping.
+### Our Proposal
+
+This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools**, basically standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors. This approach benefits both direct accessibility (helping users with disabilities navigate websites) and indirect accessibility (enabling AI assistive technologies to work reliably).
+
+Instead of agents saying "find the element with class `.contact-info`", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating reliable digital experiences for users who depend on assistive technologies and AI agents for web navigation.
 
 ## Current State: The API Dependency Challenge
 
@@ -59,11 +60,11 @@ When APIs aren't available, AI agents become digital archaeologists, excavating
 
 - **CSS Selector Archaeology**: Hunting for `.contact-info` classes that vanish with the next design update
 - **HTML Structure Divination**: Parsing unstructured content that changes without notice  
-- **Site-Specific Scripting**: Maintaining individual automation for each website—a maintenance nightmare
+- **Site-Specific Scripting**: Maintaining individual automation for each website which is a maintenance nightmare
 
 ### Real-World Pain Points
 
-Consider an AI assistant tasked with "auditing accessibility statements across our company's website portfolio." For sites with APIs, this is straightforward. For the majority without APIs, the agent must:
+Consider an AI assistant tasked with "auditing accessibility statements across a company's website portfolio." For sites with APIs, this is straightforward. For the majority without APIs, the agent must:
 
 1. **Guess** where accessibility statements might be located
 2. **Hope** the HTML structure hasn't changed since the last update
@@ -114,23 +115,23 @@ The primary approach for sensitive operations should be human-in-the-loop, where
 
 ## Real-World Applications
 
-Imagine the possibilities when AI agents can reliably navigate any website using semantic markers:
+Imagine the possibilities when AI agents can reliably navigate any website using semantic markers, particularly for users who face barriers in traditional web navigation:
 
-**The Accessibility Auditor**: A compliance officer asks their AI assistant to "check accessibility statements across our 50 partner websites." Instead of manually visiting each site or maintaining fragile scrapers, the AI agent systematically discovers and analyzes `accessibility-statement` destinations across all sites in minutes.
+**The Accessibility Compliance Assistant**: A disability rights advocate asks their AI assistant to "check accessibility statements across our 50 partner websites." Instead of manually visiting each site or maintaining fragile scrapers, the AI agent systematically discovers and analyzes `accessibility-statement` destinations across all sites in minutes, ensuring comprehensive accessibility monitoring.
 
-**The Support Research Assistant**: A customer facing issues with multiple service providers asks for help options. The AI agent discovers `contact` and `help` destinations across different platforms, presenting a comprehensive support landscape without breaking when sites redesign.
+**The Cognitive Support Assistant**: A user with cognitive disabilities facing issues with multiple service providers asks for help options. The AI agent discovers `contact` and `help` destinations across different platforms, presenting a comprehensive support landscape in a simplified, consistent format without breaking when sites redesign.
 
-**The Digital Estate Manager**: A user wants to update their profile information across platforms. The AI agent navigates to `account-settings` destinations on each site, then guides the user through the updates with full context and control.
+**The Motor Accessibility Aid**: A user with motor impairments wants to update their profile information across platforms but finds repeated navigation challenging. The AI agent navigates to `account-settings` destinations on each site, then guides the user through the updates with full context and control, reducing the physical navigation burden.
 
-**The Compliance Monitor**: A legal team needs regular checks of privacy policies across their digital ecosystem. Semantic destinations enable automated discovery and monitoring without the maintenance burden of traditional scraping.
+**The Accessibility Monitoring System**: Disability services organizations need regular checks of accessibility resources across their digital ecosystem. Semantic destinations enable automated discovery of `accessibility-statement` and `help` destinations without the maintenance burden of traditional scraping, ensuring accessibility information remains current and accessible.
 
-These scenarios share a common pattern: **AI agents handle the discovery and navigation complexity, while humans maintain control over sensitive decisions and actions.**
+These scenarios share a common pattern: **AI agents handle the discovery and navigation complexity, while humans maintain control over sensitive decisions and actions, this is particularly important for users who may need additional time or alternative interaction methods.**
 
 ## Building on Well-known Destinations
 
-The beauty of this approach lies in its foundation: the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need.
+The beauty of this approach lies in its foundation: the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need, with a core focus on accessibility.
 
-**From Human Accessibility to AI Navigation**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents—what helps humans navigate also helps machines understand.
+**From Human Accessibility to AI-Assisted Accessibility**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents acting as assistive technologies—what helps humans with disabilities navigate also enables machines to provide consistent assistance.
 
 **The Technical Foundation**: Websites already mark important destinations using standard HTML:
 
@@ -142,32 +143,34 @@ The beauty of this approach lies in its foundation: the [Well-known Destinations
 
 **Semantic Discovery**: Instead of guessing where functionality might be, AI agents can programmatically discover what each website offers through these semantic markers. This transforms unpredictable navigation into reliable discovery.
 
-**Progressive Enhancement**: Sites don't need to rebuild anything—they simply add semantic markers to their existing pages. This creates a path for gradual ecosystem adoption without disrupting current workflows.
+**Progressive Enhancement**: Sites don't need to rebuild anything, they simply add semantic markers to their existing pages. This creates a path for gradual ecosystem adoption without disrupting current workflows.
 
 ## How It Works: The Technical Foundation
 
 The architecture is elegantly simple, built on three core capabilities:
 
-**1. Destination Discovery**: AI agents enumerate available destinations by reading the semantic markers websites provide—either through HTML `<link>` elements or centralized linkset documents.
+**1. Destination Discovery**: AI agents enumerate available destinations by reading the semantic markers websites provide, either through HTML `<link>` elements or centralized linkset documents.
 
-**2. Reliable Navigation**: Using discovered destinations, agents convert semantic names like `contact` into actual URLs and navigate confidently to the right pages.
+**2. Reliable Navigation**: Using discovered destinations, agents resolve semantic names like `contact` into actual URLs and navigate confidently to the right pages.
 
 **3. Intelligent Content Processing**: Once at the destination, agents retrieve page content and hand it to Large Language Models (LLMs) for intelligent extraction and analysis.
 
 This division creates a **separation of concerns**: simple tools handle the mechanical navigation, while sophisticated AI handles the intelligent understanding of content.
 
-### The AI Integration Story
+### AI System Integration Through MCP
 
-The magic happens in how Semantic Web Tools integrate with AI systems through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification)—an open standard that allows AI applications to securely connect with external tools.
+Semantic Web Tools integrate with AI systems through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification), an open standard that enables AI applications to securely connect with external tools and data sources.
 
-**The Workflow**: When a user asks an AI agent to "find support options across my service providers," here's what happens:
+**Integration Workflow**: When a user requests an AI agent to "find support options across my service providers," the system follows this pattern:
 
-1. **Discovery Phase**: The agent uses Semantic Web Tools to discover what destinations each website offers
-2. **Planning Phase**: Based on available destinations, the agent plans the optimal navigation strategy  
+1. **Discovery Phase**: The agent uses Semantic Web Tools to discover available destinations on each target website
+2. **Planning Phase**: Based on discovered destinations, the agent determines the optimal navigation strategy  
 3. **Execution Phase**: The agent navigates to relevant destinations and retrieves content
-4. **Analysis Phase**: AI processes the retrieved content to extract and synthesize useful information
+4. **Analysis Phase**: The AI processes the retrieved content to extract and synthesize information
 
-This pattern works consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction.
+**Coordination Model**: This workflow emerges from LLM reasoning capabilities rather than pre-programmed algorithms. The AI agent uses language understanding to dynamically interpret requirements, assess available destinations, and adapt its approach based on what it discovers. The semantic foundation provided by Well-known Destinations enables this flexible, reasoning-based coordination to operate reliably across different websites.
+
+This integration pattern works consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction.
 
 ## Semantic Web Tools Architecture
 
@@ -177,7 +180,7 @@ Semantic Web Tools provide standardized capabilities that work uniformly across
 
 #### Core Capabilities
 
-* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site through parsing `<link>` elements or accessing linkset documents
+* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site through parsing `<link>` or `<a>` elements or accessing linkset documents
 * **Semantic Navigation**: Navigating to specific destinations using semantic identifiers rather than site-specific selectors
 * **Content Retrieval**: Fetching page content from destination URLs and delivering it to LLMs for intelligent processing
 
@@ -245,7 +248,7 @@ The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/speci
 #### MCP Architecture Overview
 
 MCP operates on a client-server model:
-- **MCP Clients**: AI applications (like Claude, ChatGPT, or custom AI agents)
+- **MCP Clients**: AI applications (agents)
 - **MCP Servers**: Applications that expose tools and resources to AI systems
 - **Protocol**: Standardized JSON-RPC communication between clients and servers
 
@@ -258,6 +261,29 @@ Semantic Web Tools can be exposed through MCP servers, providing AI agents with
 * **Standardized Interface**: AI agents can interact with Semantic Web Tools using standard MCP protocols
 * **Ecosystem**: Leverage the growing MCP ecosystem of tools and integrations
 
+#### WebMCP vs Traditional MCP Server Deployment
+
+**What is WebMCP?**
+WebMCP is a browser-based implementation of MCP that allows websites to expose tools directly through embedded JavaScript widgets. The website itself becomes an MCP server endpoint.
+
+**What is Traditional MCP Server?**
+Traditional MCP servers are standalone applications that implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website URL as a parameter.
+
+**Both Approaches Support Well-known Destinations Standardization:**
+Since Well-known Destinations provides standardized discovery and navigation, both deployment models can implement the same standardized tools:
+
+**Traditional MCP Server Approach:**
+- Single server implements standardized tools like `discoverDestinations(url)`
+- Tools are centralized but work across any compliant website
+- Server handles the standardized discovery and navigation logic
+- Easier to maintain consistency across the tool implementations
+
+**WebMCP Approach:**
+- Each website implements the same standardized tools locally
+- Tools are distributed but follow the same semantic web standards
+- Websites expose their own Well-known Destinations through embedded widgets
+- Lower deployment barrier for individual website adoption
+
 ## Agentic AI Integration Patterns
 
 Agentic AI systems integrate with Semantic Web Tools through a straightforward workflow that leverages semantic destinations for reliable website interaction. The integration follows a consistent pattern regardless of the specific task or websites involved.
@@ -288,6 +314,27 @@ AI agents process the retrieved content using LLMs to extract relevant informati
 
 This integration pattern enables AI agents to work reliably across any collection of websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
 
+### Semantic Web Tools API Specification
+
+Semantic Web Tools expose a standardized set of capabilities through MCP that work uniformly across any website implementing Well-known Destinations. Below are the sample tool signatures:
+
+#### Sample Semantic Web Tools
+
+```
+Tool: discoverDestinations
+Description: Discover available semantic destinations on a page
+Parameters:
+- url (string): Page URL to discover destinations from
+Returns: List of available destinations with types and URLs
+
+Tool: navigateToDestination  
+Description: Navigate to a semantic destination and retrieve page content
+Parameters:
+- url (string): Base page URL containing the destination reference
+- destinationType (string): Destination type (e.g., contact, help, accessibility-statement)
+Returns: Navigation result with destination URL and content
+```
+
 ## Division of Responsibilities: LLM vs Semantic Web Tools
 
 The system operates through a clear separation of concerns between the LLM (reasoning/coordination) and Semantic Web Tools (technical execution):
@@ -316,16 +363,17 @@ This division ensures that the intelligent processing of unstructured web conten
 
 ## Extended Destination Types for AI Tools
 
-The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating where APIs, schemas, and webhooks don't exist, a few additional destination types could be beneficial. These are examples of potential extensions, as more use cases may be discovered in the future.
+The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating as assistive technologies where APIs, schemas, and webhooks don't exist, a few additional destination types could be beneficial for accessibility support. These are examples of potential extensions that would benefit both direct accessibility and AI-assisted accessibility.
 
 ### Example Additional Destinations
 
-* **`search`**: Site-specific search functionality when no search API is available
-* **`account-settings`**: User profile/account management pages
+* **`search`**: Site-specific search functionality when no search API is available—critical for users who rely on AI assistance to find information
+* **`account-settings`**: User profile/account management pages—essential for users who need AI assistance managing their digital presence
+* **`accessibility-preferences`**: User accessibility settings and preferences—enabling AI agents to help users configure accessible experiences
 
-These destination types address common agentic AI use cases that cannot be handled through existing Well-known Destinations and where sites lack APIs. Importantly, these additional destinations would also benefit human users by providing clearer navigation paths to search functionality and account management.
+These destination types address common accessibility and agentic AI use cases that cannot be handled through existing Well-known Destinations and where sites lack APIs. Importantly, these additional destinations would directly benefit users with disabilities by providing clearer navigation paths to critical functionality while enabling AI assistive technologies to operate reliably.
 
-As the ecosystem evolves and more agentic AI use cases emerge, additional destination types may be identified and proposed through the standard development process.
+As the ecosystem evolves and more accessibility-focused agentic AI use cases emerge, additional destination types may be identified and proposed through the standard development process.
 
 ## Open Questions
 

From e839dea05bb79956a21cdd3db90bbe995ebd48f9 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Tue, 21 Oct 2025 18:56:07 +0530
Subject: [PATCH 06/13] Refine terminology and examples in explainer for
 clarity and consistency

---
 ...web-tools-using-well-known-destinations.md | 48 +++++++++----------
 1 file changed, 24 insertions(+), 24 deletions(-)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index e7f0bd6..6c4d39e 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -21,7 +21,7 @@
 - [Semantic Web Tools Architecture](#semantic-web-tools-architecture)
   * [Semantic Web Tools Capabilities](#semantic-web-tools-capabilities)
   * [Discovery and Navigation Mechanisms](#discovery-and-navigation-mechanisms)
-  * [MCP Server Integration](#mcp-server-integration)
+  * [Example Integration: MCP Servers](#example-integration-mcp-servers)
 - [Agentic AI Integration Patterns](#agentic-ai-integration-patterns)
 - [Division of Responsibilities: LLM vs Semantic Web Tools](#division-of-responsibilities-llm-vs-semantic-web-tools)
 - [Extended Destination Types for AI Tools](#extended-destination-types-for-ai-tools)
@@ -157,11 +157,11 @@ The architecture is elegantly simple, built on three core capabilities:
 
 This division creates a **separation of concerns**: simple tools handle the mechanical navigation, while sophisticated AI handles the intelligent understanding of content.
 
-### AI System Integration Through MCP
+### Example: AI System Integration Through MCP
 
-Semantic Web Tools integrate with AI systems through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification), an open standard that enables AI applications to securely connect with external tools and data sources.
+One way Semantic Web Tools can integrate with AI systems is through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification), an open standard that enables AI applications to securely connect with external tools and data sources. This represents one example of how such integration could work.
 
-**Integration Workflow**: When a user requests an AI agent to "find support options across my service providers," the system follows this pattern:
+**Example Integration Workflow**: When a user requests an AI agent to "find support options across my service providers," a system using MCP might follow this pattern:
 
 1. **Discovery Phase**: The agent uses Semantic Web Tools to discover available destinations on each target website
 2. **Planning Phase**: Based on discovered destinations, the agent determines the optimal navigation strategy  
@@ -170,7 +170,7 @@ Semantic Web Tools integrate with AI systems through the [Model Context Protocol
 
 **Coordination Model**: This workflow emerges from LLM reasoning capabilities rather than pre-programmed algorithms. The AI agent uses language understanding to dynamically interpret requirements, assess available destinations, and adapt its approach based on what it discovers. The semantic foundation provided by Well-known Destinations enables this flexible, reasoning-based coordination to operate reliably across different websites.
 
-This integration pattern works consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction.
+This example integration pattern demonstrates how semantic web tools could work consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction. Other integration approaches and protocols could achieve similar results.
 
 ## Semantic Web Tools Architecture
 
@@ -241,9 +241,9 @@ GET /.well-known/destinations
 
 This would return a standardized JSON format describing available destinations and any additional metadata needed for automated processing.
 
-### MCP Server Integration
+### Example Integration: MCP Servers
 
-The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification) is an open standard that enables AI applications to securely connect to external systems and data sources. MCP defines a standardized way for AI applications to interact with tools, resources, and services through a simple protocol.
+The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification) provides one example of how semantic web tools could integrate with AI systems. MCP is an open standard that enables AI applications to securely connect to external systems and data sources, defining a standardized way for AI applications to interact with tools, resources, and services through a simple protocol.
 
 #### MCP Architecture Overview
 
@@ -256,37 +256,37 @@ MCP operates on a client-server model:
 
 Semantic Web Tools can be exposed through MCP servers, providing AI agents with standardized access to website interaction capabilities. This integration allows AI systems to discover destinations, navigate to semantic endpoints, and retrieve content for LLM processing through the established MCP protocol.
 
-#### Benefits of MCP Integration
+#### Benefits of This Integration Approach
 
-* **Standardized Interface**: AI agents can interact with Semantic Web Tools using standard MCP protocols
-* **Ecosystem**: Leverage the growing MCP ecosystem of tools and integrations
+* **Standardized Interface**: AI agents can interact with Semantic Web Tools using established protocols like MCP
+* **Ecosystem Leverage**: Can take advantage of existing tool ecosystems and integrations
+* **Flexibility**: Other integration patterns and protocols could provide similar benefits
 
-#### WebMCP vs Traditional MCP Server Deployment
+#### Example Deployment Patterns with MCP
 
-**What is WebMCP?**
-WebMCP is a browser-based implementation of MCP that allows websites to expose tools directly through embedded JavaScript widgets. The website itself becomes an MCP server endpoint.
+**Browser-based Integration (WebMCP Example):**
+One approach allows websites to expose tools directly through embedded JavaScript , where the website itself becomes a tool server endpoint.
 
-**What is Traditional MCP Server?**
-Traditional MCP servers are standalone applications that implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website URL as a parameter.
+**Standalone Server Integration (Traditional MCP Example):**
+Another approach uses standalone applications that implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website page URL as a parameter.
 
-**Both Approaches Support Well-known Destinations Standardization:**
-Since Well-known Destinations provides standardized discovery and navigation, both deployment models can implement the same standardized tools:
+**Both Example Approaches Support Well-known Destinations:**
+Since Well-known Destinations provides standardized discovery and navigation, different deployment models can implement the same standardized tools:
 
-**Traditional MCP Server Approach:**
+**Standalone Server Example:**
 - Single server implements standardized tools like `discoverDestinations(url)`
 - Tools are centralized but work across any compliant website
 - Server handles the standardized discovery and navigation logic
 - Easier to maintain consistency across the tool implementations
 
-**WebMCP Approach:**
+**Browser-based Integration Example:**
 - Each website implements the same standardized tools locally
 - Tools are distributed but follow the same semantic web standards
-- Websites expose their own Well-known Destinations through embedded widgets
-- Lower deployment barrier for individual website adoption
+
 
 ## Agentic AI Integration Patterns
 
-Agentic AI systems integrate with Semantic Web Tools through a straightforward workflow that leverages semantic destinations for reliable website interaction. The integration follows a consistent pattern regardless of the specific task or websites involved.
+Agentic AI systems can integrate with Semantic Web Tools through various approaches that leverage semantic destinations for reliable website interaction. While specific integration methods may vary, the core workflow pattern remains consistent regardless of the specific task or websites involved.
 
 ### Core Integration Workflow
 
@@ -314,9 +314,9 @@ AI agents process the retrieved content using LLMs to extract relevant informati
 
 This integration pattern enables AI agents to work reliably across any collection of websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
 
-### Semantic Web Tools API Specification
+### Example: Semantic Web Tools API Specification
 
-Semantic Web Tools expose a standardized set of capabilities through MCP that work uniformly across any website implementing Well-known Destinations. Below are the sample tool signatures:
+Semantic Web Tools could expose a standardized set of capabilities that work uniformly across any website implementing Well-known Destinations. The following shows example tool signatures that could be implemented through various integration approaches (such as MCP or other protocols):
 
 #### Sample Semantic Web Tools
 

From 90fdf6c6eb6ac8bc7432cee2dcd6cde4f8b58825 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Tue, 21 Oct 2025 19:22:46 +0530
Subject: [PATCH 07/13] Add section on AI agents' integration with semantic
 destinations

---
 ...antic-web-tools-using-well-known-destinations.md | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
index 6c4d39e..e7721b3 100644
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ b/explainers/semantic-web-tools-using-well-known-destinations.md
@@ -145,6 +145,19 @@ The beauty of this approach lies in its foundation: the [Well-known Destinations
 
 **Progressive Enhancement**: Sites don't need to rebuild anything, they simply add semantic markers to their existing pages. This creates a path for gradual ecosystem adoption without disrupting current workflows.
 
+## How AI Agents Learn to Use Semantic Destinations
+
+A key question emerges: **How do AI agents actually learn to use semantic markers like `rel="contact"`?** The practical answer is through explicit tool integration rather than expecting agents to discover these patterns naturally.
+
+### Explicit Tool Integration
+
+AI agents don't need to "learn" semantic markers through training data. Instead, they are explicitly equipped with Semantic Web Tools that understand Well-known Destinations:
+
+**Active Tool Integration**: AI frameworks like MCP (Model Context Protocol) allow developers to provide agents with specific tools that know how to discover and navigate semantic destinations. The agent doesn't need to "learn" that `rel="contact"` means contact information, it uses a tool that's explicitly programmed with this knowledge.
+
+**Instruction-Following**: Modern LLMs excel at following instructions. When given tools that can "find destinations on this website" or "navigate to the contact page," they can coordinate these capabilities effectively without needing prior training on semantic web standards.
+
+
 ## How It Works: The Technical Foundation
 
 The architecture is elegantly simple, built on three core capabilities:

From c8e4424ecfb56f9564df2da29ca22145df2aa568 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Wed, 22 Oct 2025 21:12:27 +0530
Subject: [PATCH 08/13] Refactored document

---
 ...or-agents-using-well-known-destinations.md | 311 +++++++++++++
 ...web-tools-using-well-known-destinations.md | 411 ------------------
 2 files changed, 311 insertions(+), 411 deletions(-)
 create mode 100644 explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
 delete mode 100644 explainers/semantic-web-tools-using-well-known-destinations.md

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
new file mode 100644
index 0000000..079fb50
--- /dev/null
+++ b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
@@ -0,0 +1,311 @@
+# WAI-Adapt: Navigation and Content Tools for Agentic AI Using Well-known Destinations Explainer
+
+## Authors
+
+- Abhinav Kumar (@abhina-kumar), SAP Labs India Pvt. Ltd.
+
+## Participate
+
+* Issues: https://github.com/w3c/adapt/issues
+* Discussions: https://github.com/w3c/adapt/discussions
+
+## Contents
+
+- [Introduction](#introduction)
+- [The Problem: API Dependency Challenge](#the-problem-api-dependency-challenge)
+- [The Solution: Semantic Web Tools](#the-solution-semantic-web-tools)
+- [Building on Well-known Destinations](#building-on-well-known-destinations)
+- [Technical Foundation](#technical-foundation)
+- [Real-World Applications](#real-world-applications)
+- [Scope and Limitations](#scope-and-limitations)
+- [Extended Destination Types for AI Tools](#extended-destination-types-for-ai-tools)
+- [Open Questions](#open-questions)
+- [References](#references)
+- [Appendix: Background on Agentic AI and Integration Patterns](#appendix-background-on-agentic-ai-and-integration-patterns)
+
+## Introduction
+
+**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps. For users with disabilities, these AI agents can serve as powerful assistive technologies, helping navigate digital environments and complete tasks that might otherwise be challenging.
+
+However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs.
+
+This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools** - standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
+
+Instead of AI agents looking for contact information using fragile selectors like "find the element with class `.contact-info`", they would use semantic identifiers to say "navigate to the 'contact' destination". This semantic approach works consistently across all compliant websites, creating reliable digital experiences.
+
+This approach benefits both direct accessibility (helping users with disabilities navigate websites) and indirect accessibility (enabling AI assistive technologies to work reliably).
+
+## The Problem: API Dependency Challenge
+
+Today's agentic AI systems face a stark reality, they live in a two-tier world of website interaction.
+
+**The Privileged Few (20-30% of sites):**
+Sites like Twitter, GitHub provide robust APIs. Here AI agents work beautifully, they can reliably access data, perform actions, and integrate seamlessly into workflows.
+
+**The Struggling Majority (70-80% of sites):**
+Most websites from local businesses to government agencies don't provide APIs for basic functions like finding contact information, accessing help resources, or managing user accounts. This forces AI agents into a world of fragile workarounds:
+
+- **CSS Selector Archaeology**: Hunting for `.contact-info` classes that vanish with the next design update
+- **HTML Structure Divination**: Parsing unstructured content that changes without notice  
+- **Site-Specific Scripting**: Maintaining individual automation for each website which is a maintenance nightmare
+
+### Real-World Pain Points
+
+Consider an AI assistant tasked with "auditing accessibility statements across a company's website portfolio." For sites with APIs, this is straightforward. For the majority without APIs, the agent must:
+
+1. Guess where accessibility statements might be located
+2. Hope the HTML structure hasn't changed since the last update
+3. Fail gracefully when selectors break, leaving users with incomplete results
+
+**The Core Problem**: We need a semantic bridge between the reliability of APIs and the universality of web interfaces.
+
+| Approach | Coverage | Reliability | Maintenance Reality |
+|----------|----------|-------------|---------------------|
+| **APIs Only** | ~20-30% of sites | High | Low (when available) |
+| **Traditional Scraping** | ~100% of sites | Very Low | Constant firefighting |
+| **Semantic Web Tools** | Growing with adoption | High | Set-and-forget |
+
+## The Solution: Semantic Web Tools
+
+Well-known Destinations offer an elegant solution to this challenge. Instead of websites needing to build full APIs, they can provide semantic signposts that AI agents can follow reliably.
+
+Rather than an AI agent hunting for contact information using fragile selectors like `document.querySelector('.contact-page-link')`, it simply looks for the Well-Known Destination for Contact page. This approach transforms brittle technical navigation into reliable semantic discovery.
+
+## Building on Well-known Destinations
+
+The beauty of this approach lies in its foundation, the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need, with a core focus on accessibility.
+
+**From Human Accessibility to AI-Assisted Accessibility**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents also acting as assistive technologies, what helps humans with disabilities navigate also enables machines to provide consistent assistance. These Well-known Destinations act as a basis for tools used by agents to discover and navigate to these destinations.
+
+## Technical Foundation
+
+The architecture is elegantly simple, built on three core capabilities:
+
+### Core Web Tools Capabilities
+
+* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site / page 
+* **Semantic Navigation**: Navigating to specific destinations using semantic identifiers rather than site-specific selectors
+* **Content Retrieval**: Fetching page content from destination URLs and delivering it to LLMs for intelligent processing
+
+**Benefits of Semantic Approach:**
+- Single set of capabilities works across all compliant websites
+- No site-specific customization or maintenance required
+- Consistent, predictable behavior for AI agents
+- Leverages semantic web standards for reliability
+
+### Discovery Mechanisms
+
+Semantic Web Tools leverage the standardized discovery mechanisms already defined in the Well-known Destinations proposal. The Well-known Destinations framework supports multiple discovery methods (like linkset), ensuring AI tools can adapt as new optimization approaches emerge.
+
+### Division of Responsibilities
+
+The system operates through a clear separation of concerns between the LLM (reasoning/coordination) and Semantic Web Tools (technical execution):
+
+**LLM Responsibilities:**
+* **Task Planning**: Interpreting user requests and determining which destinations are needed from available destinations.
+* **Workflow Coordination**: Orchestrating the sequence of operations
+* **Web Page Analysis**: Synthesizing information from web page into coherent responses (like getting a hotel's reception number)
+* **Error Handling**: Deciding how to respond when tools encounter failures or missing content
+* **User Interaction**: Managing clarifications, progress updates, and result presentation
+
+**Web Tools Responsibilities:**
+* **Destination Discovery**: Finding available Well-known Destinations on target websites
+* **Navigation**: Converting semantic destination names to URLs and performing HTTP requests
+* **Content Retrieval**: Fetching raw page content and passing it to LLMs for processing
+* **Error Reporting**: Detecting and reporting technical failures (network errors, missing pages, authentication failures) to the LLM for decision-making
+
+### Content Processing Strategy
+
+Semantic Web Tools handle only the technical aspects of content retrieval, while LLMs handle all content processing:
+
+- **Semantic Web Tools**: Navigate to destination URLs or fetch raw HTML content from destination
+- **LLMs**: Extract, clean, structure, and interpret the content based on specific task requirements
+
+This division ensures that intelligent processing of unstructured web content is handled by LLMs, which can adapt to different page layouts, content types, and extraction requirements. Technical issues like rate limiting, authentication failures, or access restrictions are reported to the LLM, which decides how to respond (retry, use alternative destinations, request human intervention, etc.).
+
+## Real-World Applications
+
+Imagine the possibilities when AI agents can reliably navigate any website using semantic markers, particularly for users who face barriers in traditional web navigation:
+
+**The Accessibility Compliance Assistant**: A disability rights advocate asks their AI assistant to "check accessibility statements across our 50 partner websites." Instead of manually visiting each site or maintaining fragile scrapers, the AI agent systematically discovers and analyzes `accessibility-statement` destinations across all sites in minutes, ensuring comprehensive accessibility monitoring.
+
+**The Cognitive Support Assistant**: A user with cognitive disabilities facing issues with multiple service providers asks for help options. The AI agent discovers `contact` and `help` destinations across different platforms, presenting a comprehensive support landscape in a simplified, consistent format without breaking when sites redesign.
+
+**The Motor Accessibility Aid**: A user with motor impairments wants to update their profile information across platforms but finds repeated navigation challenging. The AI agent navigates to `account-settings` destinations on each site, then guides the user through the updates with full context and control, reducing the physical navigation burden.
+
+**The Accessibility Monitoring System**: Disability services organizations need regular checks of accessibility resources across their digital ecosystem. Semantic destinations enable automated discovery of `accessibility-statement` and `help` destinations without the maintenance burden of traditional scraping, ensuring accessibility information remains current and accessible.
+
+These scenarios share a common pattern: AI agents handle the discovery and navigation complexity using LLMs for planning and coordination, while Semantic Web Tools provide reliable technical execution. Humans maintain control over sensitive decisions and actions.
+
+## Scope and Limitations
+
+### What Well-known Destinations Solve Well
+
+* **Content Discovery**: Finding specific types of pages (help, contact, accessibility statements, product catalogs)
+* **Information Extraction**: Extracting data from destination pages
+* **Consistent Navigation**: Standardized navigation patterns across different websites
+
+### What Requires Additional Solutions
+
+* **Complex Actions**: Multi-step workflows like password changes, account modifications, or transaction processing
+  - *Why Well-known Destinations aren't sufficient*: While destinations can navigate to account settings pages, the actual modification workflows vary dramatically between sites. A password change might require current password verification, security questions, or two-factor authentication in site-specific ways that can't be standardized through destinations alone.
+  - *Solution approach*: Well-known Destinations provide navigation to relevant areas, then additional tools (APIs, human-in-the-loop) handle the complex workflows.
+
+* **Authenticated Operations**: Actions requiring user-specific authentication and authorization  
+  - *Why Well-known Destinations aren't sufficient*: While an agent could navigate to a login page using destinations, the authentication mechanisms (OAuth, SAML, multi-factor authentication) require site-specific integration that goes beyond semantic navigation.
+  - *Solution approach*: Destinations can navigate to login/authentication pages, but secure authentication requires specialized protocols and often human involvement for security.
+
+* **Site-specific Workflows**: Unique business processes that don't map to standard destinations
+  - *Why Well-known Destinations aren't sufficient*: Business processes like "submit insurance claim" or "schedule medical appointment" involve complex, industry-specific workflows that can't be standardized across all websites.
+  - *Solution approach*: Destinations help reach relevant sections, but specialized tools or APIs are needed for complex business logic.
+
+### Human-in-the-Loop for Complex Scenarios
+
+For complex scenarios, a layered approach works best:
+
+1. **Well-known Destinations**: Provide navigation to relevant areas
+2. **API Integration**: Use APIs where available for sensitive operations
+3. **Human-in-the-Loop**: Include human oversight for sensitive or complex tasks
+
+The primary approach for sensitive operations should be human-in-the-loop, where AI agents navigate to relevant pages using semantic destinations, extract and present available options to users, then hand control to humans for actual execution.
+
+## Extended Destination Types for AI Tools
+
+The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating as assistive technologies, a few additional destination types could be beneficial:
+
+### Example Additional Destinations
+
+* **`search`**: Site-specific search functionality when no search API is available, critical for users who rely on AI assistance to find information
+* **`account-settings`**: User profile/account management pages, essential for users who need AI assistance managing their digital presence
+* **`accessibility-preferences`**: User accessibility settings and preferences, enabling AI agents to help users configure accessible experiences
+
+These destination types address common accessibility and agentic AI use cases where sites lack APIs, directly benefiting users by providing clearer navigation paths while enabling AI assistive technologies to operate reliably.
+
+## Open Questions
+
+### Well-known Destinations Extensions
+
+* **New Destination Types**: Should additional destinations like `search` and `account-settings` be standardized?
+* **Discovery Optimization**: Should Well-known Destinations include a centralized discovery endpoint (e.g., `/.well-known/destinations`) to improve efficiency?
+
+### Standardization Process
+
+* **Governance Model**: How should new AI-relevant destination types be proposed, reviewed, and added to the standard?
+
+## References
+
+### Foundational Work
+
+* [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
+* [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
+* [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
+* [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
+
+---
+
+## Appendix: Background on Agentic AI and Integration Patterns
+
+### What Are Agentic AI Systems?
+
+**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps. For users with disabilities, these AI agents can serve as powerful assistive technologies, helping navigate digital environments and complete tasks that might otherwise be challenging.
+
+### How AI Agents Learn to Use Semantic Destinations
+
+A key question emerges: How do AI agents actually learn to use semantic markers exposed via Well-Known Destinations? The practical answer is through explicit tool integration rather than expecting agents to discover these patterns naturally.
+
+#### Explicit Tool Integration
+
+AI agents don't need to "learn" semantic markers through training data. Instead, they are explicitly equipped with Semantic Tools that understand Well-known Destinations:
+
+**Tool Integration**: AI frameworks like MCP (Model Context Protocol) allow developers to provide agents with specific tools. Such tools can be used to discover destinations, navigate to destinations, or fetch content from destinations using Well-Known Destinations.
+
+**Instruction-Following**: Modern LLMs excel at following instructions. When given tools that can "find destinations on this website" or "navigate to the contact page," they can coordinate these capabilities effectively without needing prior training.
+
+### Integration Architecture Patterns
+
+#### Core Integration Workflow
+
+Agentic AI systems integrate with Semantic Web Tools through a standardized workflow pattern that works consistently across all websites implementing Well-known Destinations:
+
+1. **Destination Discovery**: AI agents use Semantic Web Tools to discover available destinations on target webpage by using well-known destinations.
+
+2. **Destination Selection and Planning**: Based on discovered destinations and user requirements, AI agents plan which destinations to visit and in what sequence to accomplish their tasks
+
+3. **Semantic Navigation and Content Retrieval**: AI agents direct tools to navigate to selected destinations and retrieve page content for analysis
+
+4. **LLM Processing and Synthesis**: AI agents process retrieved content using LLMs to extract relevant information, make decisions, and synthesize results across multiple sites
+
+### Workflow Characteristics
+
+**Uniform Approach**: The same Semantic Web Tool and workflow pattern works across all websites that implement Well-known Destinations, eliminating the need for site-specific customization.
+
+**Semantic Reliability**: Navigation based on semantic destinations (`contact`, `help`, `accessibility-statement`) rather than fragile selectors ensures consistent operation even when sites update their designs.
+
+**Clear Separation**: Semantic Web Tools handle technical navigation and content retrieval, while LLMs handle intelligent processing and decision-making.
+
+This integration pattern enables AI agents to work reliably across any websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
+
+#### Example Integration with Model Context Protocol (MCP)
+
+The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification) provides one example of how tools could integrate with AI systems. MCP is an open standard that enables AI applications (Agents) to securely connect with external tools and data sources.
+
+**MCP Architecture Overview:**
+MCP operates on a client-server model:
+- **MCP Clients**: AI applications (agents)
+- **MCP Servers**: Applications that expose tools and resources to AI systems
+- **Protocol**: JSON-RPC over stdio, HTTP with SSE (Server-Sent Events), or WebSocket connections
+
+**Semantic Web Tools as MCP Tools:**
+Semantic Web Tools can be exposed through MCP servers, providing AI agents with standardized tools to access website.
+
+**Example MCP Integration Workflow:**
+When a user requests an AI agent to "find support options across my service providers," a system using MCP might follow this pattern:
+
+1. The agent uses Semantic Web Tools to discover available destinations on each target website
+2. Based on discovered destinations, the agent determines the optimal navigation strategy  
+3. The agent navigates to relevant destinations and retrieves content
+4. The AI processes the retrieved content to extract and synthesize information
+
+#### Example Deployment Patterns with MCP
+
+**Websites acting as MCP Server (WebMCP):**
+Websites can expose tools directly through embedded JavaScript, where the website itself becomes a tool server.
+
+**Standalone Server Integration (Traditional MCP Example):**
+Standalone applications implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website/webpage URL as a parameter.
+
+**Both approaches support Well-known Destinations** since the standard provides unified discovery and navigation that works regardless of deployment model.
+
+#### Sample Semantic Web Tools API
+
+Semantic Web Tools expose a standardized set of capabilities that work uniformly across any website implementing Well-known Destinations. These tools support two distinct workflows:
+
+**Workflow 1 - Direct Navigation**: User wants to visit a specific page (e.g., "take me to the help page")
+
+**Workflow 2 - Information Extraction**: User wants specific information (e.g., "get the customer service phone number")
+
+```
+Tool: discoverDestinations
+Description: Discover available semantic destinations on a page
+Parameters:
+- url (string): Page URL to discover destinations from
+Returns: List of available destinations with types and URLs
+
+Tool: navigateToDestination  
+Description: Navigate to a semantic destination
+Parameters:
+- url (string): Base page URL containing the destination reference
+- destinationType (string): Destination type (e.g., contact, help, accessibility-statement)
+Returns: Navigation result with destination URL
+
+Tool: fetchContentFromDestination
+Description: Navigate to a semantic destination and retrieve page content for LLM processing
+Parameters:
+- url (string): Base page URL containing the destination reference  
+- destinationType (string): Destination type (e.g., contact, help, accessibility-statement)
+Returns: Navigation result with destination URL and extracted content for LLM analysis
+```
+
+**Usage Examples:**
+- **Direct Navigation**: `navigateToDestination` for "take me to the contact page"
+- **Information Extraction**: `fetchContentFromDestination` for "get the customer service phone number" (LLM processes the content to extract specific information)
diff --git a/explainers/semantic-web-tools-using-well-known-destinations.md b/explainers/semantic-web-tools-using-well-known-destinations.md
deleted file mode 100644
index e7721b3..0000000
--- a/explainers/semantic-web-tools-using-well-known-destinations.md
+++ /dev/null
@@ -1,411 +0,0 @@
-# WAI-Adapt: Semantic Web Tools for Agentic AI Using Well-known Destinations Explainer
-
-## Authors
-
-- Abhinav Kumar (@abhina-kumar), SAP Labs India Pvt. Ltd.
-
-## Participate
-
-* Issues: https://github.com/w3c/adapt/issues
-* Discussions: https://github.com/w3c/adapt/discussions
-
-## Contents
-
-- [Introduction](#introduction)
-- [Current State: The API Dependency Challenge](#current-state-the-api-dependency-challenge)
-- [The Semantic Web Tools Solution](#the-semantic-web-tools-solution)
-- [Scope and Limitations](#scope-and-limitations)
-- [Real-World Applications](#real-world-applications)
-- [Building on Well-known Destinations](#building-on-well-known-destinations)
-- [How It Works: The Technical Foundation](#how-it-works-the-technical-foundation)
-- [Semantic Web Tools Architecture](#semantic-web-tools-architecture)
-  * [Semantic Web Tools Capabilities](#semantic-web-tools-capabilities)
-  * [Discovery and Navigation Mechanisms](#discovery-and-navigation-mechanisms)
-  * [Example Integration: MCP Servers](#example-integration-mcp-servers)
-- [Agentic AI Integration Patterns](#agentic-ai-integration-patterns)
-- [Division of Responsibilities: LLM vs Semantic Web Tools](#division-of-responsibilities-llm-vs-semantic-web-tools)
-- [Extended Destination Types for AI Tools](#extended-destination-types-for-ai-tools)
-- [Open Questions](#open-questions)
-- [References](#references)
-
-## Introduction
-
-**Agentic AI systems** are autonomous software agents that can plan, reason, and execute complex multi-step tasks on behalf of users. These systems represent a new frontier in automation, capable of understanding natural language requests and breaking them down into actionable steps. For users with disabilities, these AI agents can serve as powerful assistive technologies, helping navigate digital environments and complete tasks that might otherwise be challenging.
-
-However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs, particularly problematic for users who depend on consistent, reliable access to digital services.
-
-### The Problem Today
-
-When a user with cognitive disabilities asks an AI agent to "find contact information across my service providers," or when someone with motor impairments needs assistance navigating multiple websites, the agent faces a dilemma. For the 20-30% of sites with APIs, it can work reliably. For the remaining 70-80%, it must resort to fragile CSS selectors and HTML parsing that requires constant maintenance, causing unreliable experiences precisely when consistency matters most.
-
-### Our Proposal
-
-This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools**, basically standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors. This approach benefits both direct accessibility (helping users with disabilities navigate websites) and indirect accessibility (enabling AI assistive technologies to work reliably).
-
-Instead of agents saying "find the element with class `.contact-info`", they would say "navigate to the 'contact' destination." This semantic approach works consistently across all compliant websites, creating reliable digital experiences for users who depend on assistive technologies and AI agents for web navigation.
-
-## Current State: The API Dependency Challenge
-
-Today's agentic AI systems face a stark reality: they live in a two-tier world of website interaction.
-
-**The Privileged Few (20-30% of sites):**
-Sites like Twitter, GitHub provide robust APIs. Here, AI agents work beautifully—they can reliably access data, perform actions, and integrate seamlessly into workflows. These sites offer the structured, programmatic access that AI systems need.
-
-**The Struggling Majority (70-80% of sites):**
-Most websites—from local businesses to government agencies—don't provide APIs for basic functions like finding contact information, accessing help resources, or managing user accounts. This forces AI agents into a world of fragile workarounds.
-
-### The Brittle Fallback Reality
-
-When APIs aren't available, AI agents become digital archaeologists, excavating information through:
-
-- **CSS Selector Archaeology**: Hunting for `.contact-info` classes that vanish with the next design update
-- **HTML Structure Divination**: Parsing unstructured content that changes without notice  
-- **Site-Specific Scripting**: Maintaining individual automation for each website which is a maintenance nightmare
-
-### Real-World Pain Points
-
-Consider an AI assistant tasked with "auditing accessibility statements across a company's website portfolio." For sites with APIs, this is straightforward. For the majority without APIs, the agent must:
-
-1. **Guess** where accessibility statements might be located
-2. **Hope** the HTML structure hasn't changed since the last update
-3. **Fail gracefully** when selectors break, leaving users with incomplete results
-
-**The Core Problem**: We need a semantic bridge between the reliability of APIs and the universality of web interfaces.
-
-| Approach | Coverage | Reliability | Maintenance Reality |
-|----------|----------|-------------|---------------------|
-| **APIs Only** | ~20-30% of sites | High | Low (when available) |
-| **Traditional Scraping** | ~100% of sites | Very Low | Constant firefighting |
-| **Semantic Web Tools** | Growing with adoption | High | Set-and-forget |
-
-## The Semantic Web Tools Solution
-
-Well-known Destinations offer an elegant solution to this challenge. Instead of websites needing to build full APIs, they can provide semantic signposts that AI agents can follow reliably.
-
-**How It Works:**
-Rather than an AI agent hunting for contact information using fragile selectors like `document.querySelector('.contact-page-link')`, it simply looks for the semantic marker: `rel="contact"`. This approach transforms brittle technical navigation into reliable semantic discovery.
-
-## Scope and Limitations
-
-### What Well-known Destinations Solve Well
-
-* **Content Discovery**: Finding specific types of pages (help, contact, accessibility statements, product catalogs etc.)
-* **Information Extraction**: Extracting data from destination pages
-* **Consistent Navigation**: Standardized navigation patterns across different websites
-
-### What Requires Additional Solutions
-
-* **Complex Actions**: Multi-step workflows like password changes, account modifications, or transaction processing
-* **Authenticated Operations**: Actions requiring user-specific authentication and authorization
-* **Site-specific Workflows**: Unique business processes that don't map to standard destinations
-
-### Human-in-the-Loop and Layered Approaches for Complex Scenarios
-
-For complex scenarios like account management, a layered approach works best:
-
-1. **Well-known Destinations**: Provide navigation to account management areas
-2. **API Integration**: Use APIs where available for sensitive operations
-3. **Human-in-the-Loop**: Include human oversight and intervention for sensitive or complex tasks
-
-The primary approach for sensitive operations should be human-in-the-loop, where AI agents:
-- Navigate to the relevant pages using semantic destinations
-- Extract and present available options to users
-- Hand control to humans for actual execution of sensitive tasks
-- Provide guidance and context to facilitate user completion
-
-## Real-World Applications
-
-Imagine the possibilities when AI agents can reliably navigate any website using semantic markers, particularly for users who face barriers in traditional web navigation:
-
-**The Accessibility Compliance Assistant**: A disability rights advocate asks their AI assistant to "check accessibility statements across our 50 partner websites." Instead of manually visiting each site or maintaining fragile scrapers, the AI agent systematically discovers and analyzes `accessibility-statement` destinations across all sites in minutes, ensuring comprehensive accessibility monitoring.
-
-**The Cognitive Support Assistant**: A user with cognitive disabilities facing issues with multiple service providers asks for help options. The AI agent discovers `contact` and `help` destinations across different platforms, presenting a comprehensive support landscape in a simplified, consistent format without breaking when sites redesign.
-
-**The Motor Accessibility Aid**: A user with motor impairments wants to update their profile information across platforms but finds repeated navigation challenging. The AI agent navigates to `account-settings` destinations on each site, then guides the user through the updates with full context and control, reducing the physical navigation burden.
-
-**The Accessibility Monitoring System**: Disability services organizations need regular checks of accessibility resources across their digital ecosystem. Semantic destinations enable automated discovery of `accessibility-statement` and `help` destinations without the maintenance burden of traditional scraping, ensuring accessibility information remains current and accessible.
-
-These scenarios share a common pattern: **AI agents handle the discovery and navigation complexity, while humans maintain control over sensitive decisions and actions, this is particularly important for users who may need additional time or alternative interaction methods.**
-
-## Building on Well-known Destinations
-
-The beauty of this approach lies in its foundation: the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need, with a core focus on accessibility.
-
-**From Human Accessibility to AI-Assisted Accessibility**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents acting as assistive technologies—what helps humans with disabilities navigate also enables machines to provide consistent assistance.
-
-**The Technical Foundation**: Websites already mark important destinations using standard HTML:
-
-```html
-<link rel="contact" href="/contact-us">
-<link rel="help" href="/support">  
-<link rel="accessibility-statement" href="/accessibility">
-```
-
-**Semantic Discovery**: Instead of guessing where functionality might be, AI agents can programmatically discover what each website offers through these semantic markers. This transforms unpredictable navigation into reliable discovery.
-
-**Progressive Enhancement**: Sites don't need to rebuild anything, they simply add semantic markers to their existing pages. This creates a path for gradual ecosystem adoption without disrupting current workflows.
-
-## How AI Agents Learn to Use Semantic Destinations
-
-A key question emerges: **How do AI agents actually learn to use semantic markers like `rel="contact"`?** The practical answer is through explicit tool integration rather than expecting agents to discover these patterns naturally.
-
-### Explicit Tool Integration
-
-AI agents don't need to "learn" semantic markers through training data. Instead, they are explicitly equipped with Semantic Web Tools that understand Well-known Destinations:
-
-**Active Tool Integration**: AI frameworks like MCP (Model Context Protocol) allow developers to provide agents with specific tools that know how to discover and navigate semantic destinations. The agent doesn't need to "learn" that `rel="contact"` means contact information, it uses a tool that's explicitly programmed with this knowledge.
-
-**Instruction-Following**: Modern LLMs excel at following instructions. When given tools that can "find destinations on this website" or "navigate to the contact page," they can coordinate these capabilities effectively without needing prior training on semantic web standards.
-
-
-## How It Works: The Technical Foundation
-
-The architecture is elegantly simple, built on three core capabilities:
-
-**1. Destination Discovery**: AI agents enumerate available destinations by reading the semantic markers websites provide, either through HTML `<link>` elements or centralized linkset documents.
-
-**2. Reliable Navigation**: Using discovered destinations, agents resolve semantic names like `contact` into actual URLs and navigate confidently to the right pages.
-
-**3. Intelligent Content Processing**: Once at the destination, agents retrieve page content and hand it to Large Language Models (LLMs) for intelligent extraction and analysis.
-
-This division creates a **separation of concerns**: simple tools handle the mechanical navigation, while sophisticated AI handles the intelligent understanding of content.
-
-### Example: AI System Integration Through MCP
-
-One way Semantic Web Tools can integrate with AI systems is through the [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification), an open standard that enables AI applications to securely connect with external tools and data sources. This represents one example of how such integration could work.
-
-**Example Integration Workflow**: When a user requests an AI agent to "find support options across my service providers," a system using MCP might follow this pattern:
-
-1. **Discovery Phase**: The agent uses Semantic Web Tools to discover available destinations on each target website
-2. **Planning Phase**: Based on discovered destinations, the agent determines the optimal navigation strategy  
-3. **Execution Phase**: The agent navigates to relevant destinations and retrieves content
-4. **Analysis Phase**: The AI processes the retrieved content to extract and synthesize information
-
-**Coordination Model**: This workflow emerges from LLM reasoning capabilities rather than pre-programmed algorithms. The AI agent uses language understanding to dynamically interpret requirements, assess available destinations, and adapt its approach based on what it discovers. The semantic foundation provided by Well-known Destinations enables this flexible, reasoning-based coordination to operate reliably across different websites.
-
-This example integration pattern demonstrates how semantic web tools could work consistently across any collection of websites that implement Well-known Destinations, creating a standardized foundation for agentic web interaction. Other integration approaches and protocols could achieve similar results.
-
-## Semantic Web Tools Architecture
-
-### Semantic Web Tools Capabilities
-
-Semantic Web Tools provide standardized capabilities that work uniformly across any website implementing Well-known Destinations:
-
-#### Core Capabilities
-
-* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site through parsing `<link>` or `<a>` elements or accessing linkset documents
-* **Semantic Navigation**: Navigating to specific destinations using semantic identifiers rather than site-specific selectors
-* **Content Retrieval**: Fetching page content from destination URLs and delivering it to LLMs for intelligent processing
-
-**Benefits of Semantic Approach:**
-- Single set of capabilities works across all compliant websites
-- No site-specific customization or maintenance is required
-- Consistent, predictable behavior for AI agents
-- Leverages semantic web standards for reliability
-
-**Note**: Content extraction and analysis is handled by LLMs, while form interactions typically require human-in-the-loop approaches for security and usability reasons.
-
-### Discovery and Navigation Mechanisms
-
-Semantic Web Tools leverage the discovery mechanisms already defined in the Well-known Destinations proposal:
-
-#### HTML Link Element Parsing (Well-known Destinations Proposal)
-
-The Well-known Destinations proposal specifies that websites mark destinations using `<link>` elements in the HTML head. Semantic Web Tools utilize this existing mechanism:
-
-```html
-<head>
-  <link rel="accessibility-statement" href="/accessibility">
-  <link rel="help" href="/support">
-  <link rel="log-in" href="/sign-in">
-  <link rel="products" href="/products">
-</head>
-```
-
-#### Centralized Linkset Discovery (Well-known Destinations Alternative)
-
-As an alternative approach explored in the Well-known Destinations proposal, sites implementing RFC 9264 Linksets can provide centralized destination discovery. Semantic Web Tools can leverage this approach when available:
-
-```json
-{
-  "linkset": [
-    {
-      "anchor": "https://example.com/",
-      "accessibility-statement": [{"href": "https://example.com/accessibility"}],
-      "help": [{"href": "https://example.com/support"}],
-      "log-in": [{"href": "https://example.com/sign-in"}]
-    }
-  ]
-}
-```
-
-**Benefits of the Linkset Approach:**
-- Single request to discover all destinations
-- Centralized management of destination mappings
-- Support for complex site hierarchies
-
-#### Future Extensions
-
-While building on the Well-known Destinations foundation, future extensions could define additional discovery mechanisms specifically optimized for automated tools:
-
-```
-GET /.well-known/destinations
-```
-
-This would return a standardized JSON format describing available destinations and any additional metadata needed for automated processing.
-
-### Example Integration: MCP Servers
-
-The [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol/specification) provides one example of how semantic web tools could integrate with AI systems. MCP is an open standard that enables AI applications to securely connect to external systems and data sources, defining a standardized way for AI applications to interact with tools, resources, and services through a simple protocol.
-
-#### MCP Architecture Overview
-
-MCP operates on a client-server model:
-- **MCP Clients**: AI applications (agents)
-- **MCP Servers**: Applications that expose tools and resources to AI systems
-- **Protocol**: Standardized JSON-RPC communication between clients and servers
-
-#### Semantic Web Tools as MCP Tools
-
-Semantic Web Tools can be exposed through MCP servers, providing AI agents with standardized access to website interaction capabilities. This integration allows AI systems to discover destinations, navigate to semantic endpoints, and retrieve content for LLM processing through the established MCP protocol.
-
-#### Benefits of This Integration Approach
-
-* **Standardized Interface**: AI agents can interact with Semantic Web Tools using established protocols like MCP
-* **Ecosystem Leverage**: Can take advantage of existing tool ecosystems and integrations
-* **Flexibility**: Other integration patterns and protocols could provide similar benefits
-
-#### Example Deployment Patterns with MCP
-
-**Browser-based Integration (WebMCP Example):**
-One approach allows websites to expose tools directly through embedded JavaScript , where the website itself becomes a tool server endpoint.
-
-**Standalone Server Integration (Traditional MCP Example):**
-Another approach uses standalone applications that implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website page URL as a parameter.
-
-**Both Example Approaches Support Well-known Destinations:**
-Since Well-known Destinations provides standardized discovery and navigation, different deployment models can implement the same standardized tools:
-
-**Standalone Server Example:**
-- Single server implements standardized tools like `discoverDestinations(url)`
-- Tools are centralized but work across any compliant website
-- Server handles the standardized discovery and navigation logic
-- Easier to maintain consistency across the tool implementations
-
-**Browser-based Integration Example:**
-- Each website implements the same standardized tools locally
-- Tools are distributed but follow the same semantic web standards
-
-
-## Agentic AI Integration Patterns
-
-Agentic AI systems can integrate with Semantic Web Tools through various approaches that leverage semantic destinations for reliable website interaction. While specific integration methods may vary, the core workflow pattern remains consistent regardless of the specific task or websites involved.
-
-### Core Integration Workflow
-
-#### 1. Destination Discovery
-AI agents use the Semantic Web Tool to discover available destinations on target websites or pages by parsing well-known destination markers (`<link>` elements, linksets, or well-known URI endpoints).
-
-#### 2. Destination Selection and Planning  
-Based on discovered destinations and user requirements, AI agents plan which destinations to visit and in what sequence to accomplish their tasks.
-
-#### 3. Semantic Navigation and Content Retrieval
-AI agents direct the Semantic Web Tool to navigate to selected destinations and retrieve page content for analysis.
-
-#### 4. LLM Processing and Synthesis
-AI agents process the retrieved content using LLMs to extract relevant information, make decisions, and synthesize results across multiple sites.
-
-### Workflow Characteristics
-
-**Uniform Approach**: The same Semantic Web Tool and workflow pattern works across all websites that implement Well-known Destinations, eliminating the need for site-specific customization.
-
-**Semantic Reliability**: Navigation based on semantic destinations (`contact`, `help`, `accessibility-statement`) rather than fragile selectors ensures consistent operation even when sites update their designs.
-
-**Clear Separation**: Semantic Web Tools handle technical navigation and content retrieval, while LLMs handle intelligent processing and decision-making.
-
-**Scalable Coordination**: AI agents can easily coordinate multi-site workflows by applying the same pattern across multiple websites simultaneously.
-
-This integration pattern enables AI agents to work reliably across any collection of websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
-
-### Example: Semantic Web Tools API Specification
-
-Semantic Web Tools could expose a standardized set of capabilities that work uniformly across any website implementing Well-known Destinations. The following shows example tool signatures that could be implemented through various integration approaches (such as MCP or other protocols):
-
-#### Sample Semantic Web Tools
-
-```
-Tool: discoverDestinations
-Description: Discover available semantic destinations on a page
-Parameters:
-- url (string): Page URL to discover destinations from
-Returns: List of available destinations with types and URLs
-
-Tool: navigateToDestination  
-Description: Navigate to a semantic destination and retrieve page content
-Parameters:
-- url (string): Base page URL containing the destination reference
-- destinationType (string): Destination type (e.g., contact, help, accessibility-statement)
-Returns: Navigation result with destination URL and content
-```
-
-## Division of Responsibilities: LLM vs Semantic Web Tools
-
-The system operates through a clear separation of concerns between the LLM (reasoning/coordination) and Semantic Web Tools (technical execution):
-
-### LLM Responsibilities
-* **Task Planning**: Interpreting user requests and determining which websites and destinations are needed
-* **Workflow Coordination**: Orchestrating the sequence of operations across multiple sites
-* **Cross-site Analysis**: Synthesizing information from multiple sources into coherent responses
-* **Error Handling**: Deciding how to respond when tools encounter failures or missing content
-* **User Interaction**: Managing clarifications, progress updates, and result presentation
-
-### Semantic Web Tools Responsibilities  
-* **Destination Discovery**: Finding available Well-known Destinations on target websites
-* **Navigation**: Converting semantic destination names to URLs and performing HTTP requests
-* **Content Retrieval**: Fetching raw page content and passing it to LLMs for processing
-* **Error Reporting**: Detecting and reporting technical failures (network errors, missing pages, authentication failures) to the LLM for decision-making
-
-### Content Processing Strategy
-
-Semantic Web Tools handle only the technical aspects of content retrieval, while LLMs handle all content processing:
-
-- **Semantic Web Tools**: Navigate to destination URLs and fetch raw HTML content
-- **LLMs**: Extract, clean, structure, and interpret the content based on the specific task requirements
-
-This division ensures that the intelligent processing of unstructured web content is handled by LLMs, which can adapt to different page layouts, content types, and extraction requirements. Protocol issues like rate limiting, authentication failures, or access restrictions are reported to the LLM, which decides how to respond (retry, use alternative destinations, request human intervention, etc.).
-
-## Extended Destination Types for AI Tools
-
-The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating as assistive technologies where APIs, schemas, and webhooks don't exist, a few additional destination types could be beneficial for accessibility support. These are examples of potential extensions that would benefit both direct accessibility and AI-assisted accessibility.
-
-### Example Additional Destinations
-
-* **`search`**: Site-specific search functionality when no search API is available—critical for users who rely on AI assistance to find information
-* **`account-settings`**: User profile/account management pages—essential for users who need AI assistance managing their digital presence
-* **`accessibility-preferences`**: User accessibility settings and preferences—enabling AI agents to help users configure accessible experiences
-
-These destination types address common accessibility and agentic AI use cases that cannot be handled through existing Well-known Destinations and where sites lack APIs. Importantly, these additional destinations would directly benefit users with disabilities by providing clearer navigation paths to critical functionality while enabling AI assistive technologies to operate reliably.
-
-As the ecosystem evolves and more accessibility-focused agentic AI use cases emerge, additional destination types may be identified and proposed through the standard development process.
-
-## Open Questions
-
-The following questions remain open for extending Well-known Destinations to support agentic AI systems:
-
-### Well-known Destinations Extensions
-
-* **New Destination Types**: Should additional destinations like `search` and `account-settings` be standardized?
-* **Discovery Optimization**: Should Well-known Destinations include a centralized discovery endpoint (e.g., `/.well-known/destinations`) to improve efficiency instead of requiring parsing of individual pages?
-
-### Standardization Process
-
-* **Governance Model**: How should new AI-relevant destination types be proposed, reviewed, and added to the standard?
-
-## References
-
-### Foundational Work
-
-* [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
-* [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
-* [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
-* [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)

From 9fea07fe261d65b994ad9ede6a11b90ce986451c Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Wed, 22 Oct 2025 22:04:46 +0530
Subject: [PATCH 09/13] Add WebMCP model details and deployment patterns to
 explainer

---
 ...or-agents-using-well-known-destinations.md | 22 ++++++++++++++-----
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
index 079fb50..13395d6 100644
--- a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
+++ b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
@@ -200,6 +200,7 @@ These destination types address common accessibility and agentic AI use cases wh
 * [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
 * [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
 * [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
+* [WebMCP](https://github.com/webmachinelearning/webmcp)
 
 ---
 
@@ -266,15 +267,24 @@ When a user requests an AI agent to "find support options across my service prov
 3. The agent navigates to relevant destinations and retrieves content
 4. The AI processes the retrieved content to extract and synthesize information
 
-#### Example Deployment Patterns with MCP
+#### Deployment Patterns with MCP: WebMCP vs. Traditional MCP Server
 
-**Websites acting as MCP Server (WebMCP):**
-Websites can expose tools directly through embedded JavaScript, where the website itself becomes a tool server.
+To address the practical challenges of browser interaction and session management, two primary architectural models emerge for deploying Semantic Web Tools:
 
-**Standalone Server Integration (Traditional MCP Example):**
-Standalone applications implement standardized semantic web tools. A single server can provide tools that work across multiple websites by taking the target website/webpage URL as a parameter.
+1.  **Traditional MCP Server**:
+    -   **How it Works**: In this model, the MCP server and its tools run on a remote server, completely independent of the user's browser. When a tool like `navigateToDestination` is invoked, the server makes a direct HTTP request to the target website.
+    -   **Strengths**: Simple to deploy for public data retrieval.
+    -   **Limitations**:
+        -   **No Session Access**: It has no access to the user's browser session, cookies, or authentication state. It cannot perform actions on behalf of a logged-in user.
+        -   **No Browser Interaction**: It cannot interact with the live browser environment, making it unsuitable for tasks that require triggering navigation or interacting with dynamic, client-side rendered applications.
 
-**Both approaches support Well-known Destinations** since the standard provides unified discovery and navigation that works regardless of deployment model.
+2.  **WebMCP**:
+    -   **How it Works**: In this model, the web page itself acts as an MCP Server. It uses a JavaScript API to define and expose "tools". An AI agent can discover and invokes these tools.
+    -   **Strengths**:
+        -   **Full Session Access**: Because the tools are part of the web page's code and execute in the browser, they automatically operate within the user's existing session, with full access to authentication state and page context.
+        -   **Live Browser Interaction**: Tools can directly manipulate the DOM and call existing JavaScript functions, allowing the agent to drive the user interface collaboratively.
+        
+-   **Recommendation**: The WebMCP model is better suited for any use case involving authenticated operations, personalized content, or complex user-specific workflows. The traditional server-side model is only suitable for public data scenarios.
 
 #### Sample Semantic Web Tools API
 

From 4a8dbc3e7755abc3ba0cb2481d3233c97db48364 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Wed, 22 Oct 2025 22:33:07 +0530
Subject: [PATCH 10/13] Remove outdated reference

---
 ...and-content-tools-for-agents-using-well-known-destinations.md | 1 -
 1 file changed, 1 deletion(-)

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
index 13395d6..981b12f 100644
--- a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
+++ b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
@@ -198,7 +198,6 @@ These destination types address common accessibility and agentic AI use cases wh
 
 * [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
 * [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
-* [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
 * [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
 * [WebMCP](https://github.com/webmachinelearning/webmcp)
 

From 0a64d33bfa025c920b32afadc8b3089ca13a3a01 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Thu, 23 Oct 2025 17:59:57 +0530
Subject: [PATCH 11/13] Minor Updates

---
 ...content-tools-for-agents-using-well-known-destinations.md | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
index 981b12f..c03cb3b 100644
--- a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
+++ b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
@@ -186,7 +186,7 @@ These destination types address common accessibility and agentic AI use cases wh
 ### Well-known Destinations Extensions
 
 * **New Destination Types**: Should additional destinations like `search` and `account-settings` be standardized?
-* **Discovery Optimization**: Should Well-known Destinations include a centralized discovery endpoint (e.g., `/.well-known/destinations`) to improve efficiency?
+* **Discovery Optimization**: The Well-Known Destinations specification considers alternative centralized discovery mechanisms, such as `/.well-known/destinations` and Linksets. Should these be formally adopted to support efficient discovery?
 
 ### Standardization Process
 
@@ -198,8 +198,9 @@ These destination types address common accessibility and agentic AI use cases wh
 
 * [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
 * [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
-* [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
 * [WebMCP](https://github.com/webmachinelearning/webmcp)
+* [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
+* [RFC 9264: Linksets](https://www.rfc-editor.org/rfc/rfc9264)
 
 ---
 

From f864350630da9a295521846f1ff05a5f9b9758b4 Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Sun, 9 Nov 2025 15:32:16 +0530
Subject: [PATCH 12/13] Update explainer to use "Discoverable Destinations"
 terminology throughout

---
 ...or-agents-using-well-known-destinations.md | 58 +++++++++----------
 1 file changed, 29 insertions(+), 29 deletions(-)

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
index c03cb3b..c32acf0 100644
--- a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
+++ b/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
@@ -1,4 +1,4 @@
-# WAI-Adapt: Navigation and Content Tools for Agentic AI Using Well-known Destinations Explainer
+# WAI-Adapt: Navigation and Content Tools for Agentic AI Using Discoverable Destinations Explainer
 
 ## Authors
 
@@ -14,7 +14,7 @@
 - [Introduction](#introduction)
 - [The Problem: API Dependency Challenge](#the-problem-api-dependency-challenge)
 - [The Solution: Semantic Web Tools](#the-solution-semantic-web-tools)
-- [Building on Well-known Destinations](#building-on-well-known-destinations)
+- [Building on Discoverable Destinations](#building-on-discoverable-destinations)
 - [Technical Foundation](#technical-foundation)
 - [Real-World Applications](#real-world-applications)
 - [Scope and Limitations](#scope-and-limitations)
@@ -29,7 +29,7 @@
 
 However, there's a fundamental challenge: while these AI agents excel at working with structured APIs, the vast majority of websites (70-80%) don't provide APIs for the functionality users actually need. This forces agents to fall back on brittle web scraping techniques that break whenever sites update their designs.
 
-This explainer proposes extending the [Well-known Destinations](well-known-destinations.md) approach to bridge this gap through **Semantic Web Tools** - standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
+This explainer proposes extending the [Discoverable Destinations](discoverable-destinations.md) approach to bridge this gap through **Semantic Web Tools** - standardized software components that enable reliable website navigation using semantic identifiers rather than fragile technical selectors.
 
 Instead of AI agents looking for contact information using fragile selectors like "find the element with class `.contact-info`", they would use semantic identifiers to say "navigate to the 'contact' destination". This semantic approach works consistently across all compliant websites, creating reliable digital experiences.
 
@@ -67,15 +67,15 @@ Consider an AI assistant tasked with "auditing accessibility statements across a
 
 ## The Solution: Semantic Web Tools
 
-Well-known Destinations offer an elegant solution to this challenge. Instead of websites needing to build full APIs, they can provide semantic signposts that AI agents can follow reliably.
+Discoverable Destinations offer an elegant solution to this challenge. Instead of websites needing to build full APIs, they can provide semantic signposts that AI agents can follow reliably.
 
-Rather than an AI agent hunting for contact information using fragile selectors like `document.querySelector('.contact-page-link')`, it simply looks for the Well-Known Destination for Contact page. This approach transforms brittle technical navigation into reliable semantic discovery.
+Rather than an AI agent hunting for contact information using fragile selectors like `document.querySelector('.contact-page-link')`, it simply looks for the Discoverable Destination for Contact page. This approach transforms brittle technical navigation into reliable semantic discovery.
 
-## Building on Well-known Destinations
+## Building on Discoverable Destinations
 
-The beauty of this approach lies in its foundation, the [Well-known Destinations](well-known-destinations.md) proposal already provides the semantic infrastructure we need, with a core focus on accessibility.
+The beauty of this approach lies in its foundation, the [Discoverable Destinations](discoverable-destinations.md) proposal already provides the semantic infrastructure we need, with a core focus on accessibility.
 
-**From Human Accessibility to AI-Assisted Accessibility**: Well-known Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents also acting as assistive technologies, what helps humans with disabilities navigate also enables machines to provide consistent assistance. These Well-known Destinations act as a basis for tools used by agents to discover and navigate to these destinations.
+**From Human Accessibility to AI-Assisted Accessibility**: Discoverable Destinations were designed to help humans with accessibility needs find important pages quickly. This same semantic approach perfectly serves AI agents also acting as assistive technologies, what helps humans with disabilities navigate also enables machines to provide consistent assistance. These Discoverable Destinations act as a basis for tools used by agents to discover and navigate to these destinations.
 
 ## Technical Foundation
 
@@ -83,7 +83,7 @@ The architecture is elegantly simple, built on three core capabilities:
 
 ### Core Web Tools Capabilities
 
-* **Destination Discovery**: Enumerating all available Well-known Destinations for a given site / page 
+* **Destination Discovery**: Enumerating all available Discoverable Destinations for a given site / page 
 * **Semantic Navigation**: Navigating to specific destinations using semantic identifiers rather than site-specific selectors
 * **Content Retrieval**: Fetching page content from destination URLs and delivering it to LLMs for intelligent processing
 
@@ -95,7 +95,7 @@ The architecture is elegantly simple, built on three core capabilities:
 
 ### Discovery Mechanisms
 
-Semantic Web Tools leverage the standardized discovery mechanisms already defined in the Well-known Destinations proposal. The Well-known Destinations framework supports multiple discovery methods (like linkset), ensuring AI tools can adapt as new optimization approaches emerge.
+Semantic Web Tools leverage the standardized discovery mechanisms already defined in the Discoverable Destinations proposal. The Discoverable Destinations framework supports multiple discovery methods (like linkset), ensuring AI tools can adapt as new optimization approaches emerge.
 
 ### Division of Responsibilities
 
@@ -109,7 +109,7 @@ The system operates through a clear separation of concerns between the LLM (reas
 * **User Interaction**: Managing clarifications, progress updates, and result presentation
 
 **Web Tools Responsibilities:**
-* **Destination Discovery**: Finding available Well-known Destinations on target websites
+* **Destination Discovery**: Finding available Discoverable Destinations on target websites
 * **Navigation**: Converting semantic destination names to URLs and performing HTTP requests
 * **Content Retrieval**: Fetching raw page content and passing it to LLMs for processing
 * **Error Reporting**: Detecting and reporting technical failures (network errors, missing pages, authentication failures) to the LLM for decision-making
@@ -139,7 +139,7 @@ These scenarios share a common pattern: AI agents handle the discovery and navig
 
 ## Scope and Limitations
 
-### What Well-known Destinations Solve Well
+### What Discoverable Destinations Solve Well
 
 * **Content Discovery**: Finding specific types of pages (help, contact, accessibility statements, product catalogs)
 * **Information Extraction**: Extracting data from destination pages
@@ -148,22 +148,22 @@ These scenarios share a common pattern: AI agents handle the discovery and navig
 ### What Requires Additional Solutions
 
 * **Complex Actions**: Multi-step workflows like password changes, account modifications, or transaction processing
-  - *Why Well-known Destinations aren't sufficient*: While destinations can navigate to account settings pages, the actual modification workflows vary dramatically between sites. A password change might require current password verification, security questions, or two-factor authentication in site-specific ways that can't be standardized through destinations alone.
-  - *Solution approach*: Well-known Destinations provide navigation to relevant areas, then additional tools (APIs, human-in-the-loop) handle the complex workflows.
+  - *Why Discoverable Destinations aren't sufficient*: While destinations can navigate to account settings pages, the actual modification workflows vary dramatically between sites. A password change might require current password verification, security questions, or two-factor authentication in site-specific ways that can't be standardized through destinations alone.
+  - *Solution approach*: Discoverable Destinations provide navigation to relevant areas, then additional tools (APIs, human-in-the-loop) handle the complex workflows.
 
 * **Authenticated Operations**: Actions requiring user-specific authentication and authorization  
-  - *Why Well-known Destinations aren't sufficient*: While an agent could navigate to a login page using destinations, the authentication mechanisms (OAuth, SAML, multi-factor authentication) require site-specific integration that goes beyond semantic navigation.
+  - *Why Discoverable Destinations aren't sufficient*: While an agent could navigate to a login page using destinations, the authentication mechanisms (OAuth, SAML, multi-factor authentication) require site-specific integration that goes beyond semantic navigation.
   - *Solution approach*: Destinations can navigate to login/authentication pages, but secure authentication requires specialized protocols and often human involvement for security.
 
 * **Site-specific Workflows**: Unique business processes that don't map to standard destinations
-  - *Why Well-known Destinations aren't sufficient*: Business processes like "submit insurance claim" or "schedule medical appointment" involve complex, industry-specific workflows that can't be standardized across all websites.
+  - *Why Discoverable Destinations aren't sufficient*: Business processes like "submit insurance claim" or "schedule medical appointment" involve complex, industry-specific workflows that can't be standardized across all websites.
   - *Solution approach*: Destinations help reach relevant sections, but specialized tools or APIs are needed for complex business logic.
 
 ### Human-in-the-Loop for Complex Scenarios
 
 For complex scenarios, a layered approach works best:
 
-1. **Well-known Destinations**: Provide navigation to relevant areas
+1. **Discoverable Destinations**: Provide navigation to relevant areas
 2. **API Integration**: Use APIs where available for sensitive operations
 3. **Human-in-the-Loop**: Include human oversight for sensitive or complex tasks
 
@@ -171,7 +171,7 @@ The primary approach for sensitive operations should be human-in-the-loop, where
 
 ## Extended Destination Types for AI Tools
 
-The Well-known Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating as assistive technologies, a few additional destination types could be beneficial:
+The Discoverable Destinations proposal primarily serves human accessibility needs. For agentic AI systems operating as assistive technologies, a few additional destination types could be beneficial:
 
 ### Example Additional Destinations
 
@@ -183,10 +183,10 @@ These destination types address common accessibility and agentic AI use cases wh
 
 ## Open Questions
 
-### Well-known Destinations Extensions
+### Discoverable Destinations Extensions
 
 * **New Destination Types**: Should additional destinations like `search` and `account-settings` be standardized?
-* **Discovery Optimization**: The Well-Known Destinations specification considers alternative centralized discovery mechanisms, such as `/.well-known/destinations` and Linksets. Should these be formally adopted to support efficient discovery?
+* **Discovery Optimization**: The Discoverable Destinations specification considers alternative centralized discovery mechanisms, such as `/.well-known/destinations` and Linksets. Should these be formally adopted to support efficient discovery?
 
 ### Standardization Process
 
@@ -196,7 +196,7 @@ These destination types address common accessibility and agentic AI use cases wh
 
 ### Foundational Work
 
-* [WAI-Adapt: Well-known Destinations Explainer](well-known-destinations.md)
+* [WAI-Adapt: Discoverable Destinations Explainer](discoverable-destinations.md)
 * [Model Context Protocol (MCP) Specification](https://github.com/modelcontextprotocol/specification)
 * [WebMCP](https://github.com/webmachinelearning/webmcp)
 * [RFC 8615: Well-Known URIs](https://datatracker.ietf.org/doc/html/rfc8615)
@@ -212,13 +212,13 @@ These destination types address common accessibility and agentic AI use cases wh
 
 ### How AI Agents Learn to Use Semantic Destinations
 
-A key question emerges: How do AI agents actually learn to use semantic markers exposed via Well-Known Destinations? The practical answer is through explicit tool integration rather than expecting agents to discover these patterns naturally.
+A key question emerges: How do AI agents actually learn to use semantic markers exposed via Discoverable Destinations? The practical answer is through explicit tool integration rather than expecting agents to discover these patterns naturally.
 
 #### Explicit Tool Integration
 
-AI agents don't need to "learn" semantic markers through training data. Instead, they are explicitly equipped with Semantic Tools that understand Well-known Destinations:
+AI agents don't need to "learn" semantic markers through training data. Instead, they are explicitly equipped with Semantic Tools that understand Discoverable Destinations:
 
-**Tool Integration**: AI frameworks like MCP (Model Context Protocol) allow developers to provide agents with specific tools. Such tools can be used to discover destinations, navigate to destinations, or fetch content from destinations using Well-Known Destinations.
+**Tool Integration**: AI frameworks like MCP (Model Context Protocol) allow developers to provide agents with specific tools. Such tools can be used to discover destinations, navigate to destinations, or fetch content from destinations using Discoverable Destinations.
 
 **Instruction-Following**: Modern LLMs excel at following instructions. When given tools that can "find destinations on this website" or "navigate to the contact page," they can coordinate these capabilities effectively without needing prior training.
 
@@ -226,9 +226,9 @@ AI agents don't need to "learn" semantic markers through training data. Instead,
 
 #### Core Integration Workflow
 
-Agentic AI systems integrate with Semantic Web Tools through a standardized workflow pattern that works consistently across all websites implementing Well-known Destinations:
+Agentic AI systems integrate with Semantic Web Tools through a standardized workflow pattern that works consistently across all websites implementing Discoverable Destinations:
 
-1. **Destination Discovery**: AI agents use Semantic Web Tools to discover available destinations on target webpage by using well-known destinations.
+1. **Destination Discovery**: AI agents use Semantic Web Tools to discover available destinations on target webpage by using discoverable destinations.
 
 2. **Destination Selection and Planning**: Based on discovered destinations and user requirements, AI agents plan which destinations to visit and in what sequence to accomplish their tasks
 
@@ -238,13 +238,13 @@ Agentic AI systems integrate with Semantic Web Tools through a standardized work
 
 ### Workflow Characteristics
 
-**Uniform Approach**: The same Semantic Web Tool and workflow pattern works across all websites that implement Well-known Destinations, eliminating the need for site-specific customization.
+**Uniform Approach**: The same Semantic Web Tool and workflow pattern works across all websites that implement Discoverable Destinations, eliminating the need for site-specific customization.
 
 **Semantic Reliability**: Navigation based on semantic destinations (`contact`, `help`, `accessibility-statement`) rather than fragile selectors ensures consistent operation even when sites update their designs.
 
 **Clear Separation**: Semantic Web Tools handle technical navigation and content retrieval, while LLMs handle intelligent processing and decision-making.
 
-This integration pattern enables AI agents to work reliably across any websites that implement Well-known Destinations, providing a standardized foundation for agentic web interaction.
+This integration pattern enables AI agents to work reliably across any websites that implement Discoverable Destinations, providing a standardized foundation for agentic web interaction.
 
 #### Example Integration with Model Context Protocol (MCP)
 
@@ -288,7 +288,7 @@ To address the practical challenges of browser interaction and session managemen
 
 #### Sample Semantic Web Tools API
 
-Semantic Web Tools expose a standardized set of capabilities that work uniformly across any website implementing Well-known Destinations. These tools support two distinct workflows:
+Semantic Web Tools expose a standardized set of capabilities that work uniformly across any website implementing Discoverable Destinations. These tools support two distinct workflows:
 
 **Workflow 1 - Direct Navigation**: User wants to visit a specific page (e.g., "take me to the help page")
 

From 38463e08d9c82cb08b56432160cc6845b6e9df4a Mon Sep 17 00:00:00 2001
From: Abhinav Kumar <abhina.kumar@sap.com>
Date: Sun, 9 Nov 2025 15:36:10 +0530
Subject: [PATCH 13/13] Renamed Explainer

---
 ...d-content-tools-for-agents-using-discoverable-destinations.md} | 0
 1 file changed, 0 insertions(+), 0 deletions(-)
 rename explainers/{navigation-and-content-tools-for-agents-using-well-known-destinations.md => navigation-and-content-tools-for-agents-using-discoverable-destinations.md} (100%)

diff --git a/explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md b/explainers/navigation-and-content-tools-for-agents-using-discoverable-destinations.md
similarity index 100%
rename from explainers/navigation-and-content-tools-for-agents-using-well-known-destinations.md
rename to explainers/navigation-and-content-tools-for-agents-using-discoverable-destinations.md