diff --git a/README.md b/README.md index 2a750ba..2f5157b 100644 --- a/README.md +++ b/README.md @@ -1,37 +1,157 @@ -# @agent-dispatch/mcp-server +

+ + + AgentDispatch + +

-MCP server exposing provider-neutral AgentDispatch tools. +

Spawn cloud agents from your AI.

-Tools: +

+ Claude Code, OpenClaw, and Hermes are great at planning. They choke when a job runs for hours.
+ @agent-dispatch/mcp-server hands them a managed cloud runtime — one MCP call, results when they land. +

-- `list_providers` -- `list_capabilities` -- `list_account_profiles` -- `spawn_cloud_agent` -- `dispatch_task` -- `get_task_status` -- `get_task_logs` -- `get_task_result` -- `cancel_task` +

+ Quick start · + What it does · + Supported clients & frameworks · + Config · + The rest of AgentDispatch +

-## Running +

+ npm + MCP + AWS Bedrock AgentCore + License + TypeScript +

-```bash -npm install @agent-dispatch/core @agent-dispatch/mcp-server @agent-dispatch/store-sqlite @agent-dispatch/adapter-aws-agentcore -npx agentdispatch-mcp --config agentdispatch.config.json +--- + +## Why this exists + +Local AI assistants plan brilliantly. They get cramped the moment work gets long: + +- A **deep-research** sweep across fifty pages of docs. +- An **account-wide audit** that has to touch every service. +- A **multi-hour** job that has no business sitting in your IDE's context window. + +Doing it inline burns context, blocks the chat, and dies the second you close the laptop. That's not what local agents are for. + +**AgentDispatch is the missing tool.** One MCP server gives your assistant a single primitive: *spawn a cloud agent with this instruction, come back later for the result.* The work runs on managed cloud compute. Status, logs, and the final output stream back through the same MCP channel. + +- ☁️ **Real cloud, not your laptop.** AWS Bedrock AgentCore today; new clouds plug in through one small adapter contract. +- ⏱️ **Built for marathons.** Sessions for stateful runs. Jobs for hours-long work. Sync for fast turn-around. Same API, three modes. +- 🔭 **Full visibility.** Status, paginated logs, results, cancellation — over MCP, the SDK, or the CLI. +- 🪶 **Boring defaults.** SQLite state. stdio transport. Zero hidden services. Runs on a laptop, in CI, on a server. + +## What it does + +A single MCP server exposes nine tools to your assistant: + +| Tool | What it does | +| --- | --- | +| `spawn_cloud_agent` | The shortcut. One `instruction`, defaults from a runtime profile, returns a `task_id`. | +| `dispatch_task` | The escape hatch. Full control over provider, capability, backend, target, and input. | +| `get_task_status` | Current status for a dispatched task. | +| `get_task_logs` | Paginated logs, cursor-based. | +| `get_task_result` | Final result payload when the task completes. | +| `cancel_task` | Cancel an in-flight task. | +| `list_providers` | Providers configured in this server. | +| `list_capabilities` | Capabilities a provider exposes (filterable). | +| `list_account_profiles` | Account profiles configured for dispatch. | + +The model side looks like this: + +```ts +spawn_cloud_agent({ instruction: "Audit our S3 buckets for public read and report findings." }) +// → { task_id: "task_...", status: "running" } + +get_task_status({ task_id: "task_..." }) +get_task_logs({ task_id: "task_...", cursor: 0 }) +get_task_result({ task_id: "task_..." }) ``` -The server loads account profiles, SQLite storage, and configured adapters from `agentdispatch.config.json`. +All tool inputs are validated with [Zod](https://zod.dev). See [`src/schemas.ts`](src/schemas.ts) for the exact shapes. -Before wiring it into an MCP client, validate that the stdio server can load config: +## Supported clients + +Anything that speaks MCP. The common ones today: + +- 🤖 **Claude Code** — add it to `~/.claude/mcp_settings.json` (or your project's `.mcp.json`). +- 🦅 **OpenClaw** — add it to your MCP server list and call `spawn_cloud_agent` from inside any task. +- 🪽 **Hermes** — same wire-up; long-running reasoning and tool-rich runs move to the cloud. +- Claude Desktop, Cursor, Continue, Goose, Zed — same JSON, different config file. + +## What the cloud agent runs + +`spawn_cloud_agent` and `dispatch_task` both accept an optional `framework` string, and runtime profiles carry a default. The value travels through to the worker — **AgentDispatch doesn't interpret it; the worker does.** That means you can wire any agent framework that your worker knows how to instantiate. + +Today the reference worker in [`worker-agentcore`](https://github.com/agent-dispatch/worker-agentcore) ships with **Strands** support. Other frameworks (your own agent loop, an OpenClaw-style runner, a Hermes-style long-context worker) become available the moment your worker handles their `framework` value. There's no allow-list in AgentDispatch itself. + +## Quick start ```bash +npm install \ + @agent-dispatch/core \ + @agent-dispatch/mcp-server \ + @agent-dispatch/store-sqlite \ + @agent-dispatch/adapter-aws-agentcore + npx agentdispatch-mcp --config agentdispatch.config.json --check ``` -Configure `defaults.runtime` for the simple agent path. Agents can then call `spawn_cloud_agent` with only an `instruction`; AgentDispatch resolves provider, account, backend, target mode, framework, and runtime tool defaults from the named runtime profile. +Wire it into your MCP client: + +```jsonc +// Claude Code, Claude Desktop, OpenClaw, Hermes, Cursor — same shape +{ + "mcpServers": { + "agentdispatch": { + "command": "npx", + "args": ["agentdispatch-mcp", "--config", "/abs/path/agentdispatch.config.json"] + } + } +} +``` + +The defaults come from your runtime profile — provider, account, backend, target mode, framework, and runtime tools are all resolved for you. From the model side, `spawn_cloud_agent({ instruction })` is the only call you need. + +## How it works + +``` + Claude Code · OpenClaw · Hermes · any MCP client + │ stdio + ▼ + ┌─────────────────────────────────────────────┐ + │ @agent-dispatch/mcp-server │ ← this repo + └─────────────────────┬───────────────────────┘ + │ + ▼ + ┌─────────────────────────────────────────────┐ + │ @agent-dispatch/core │ ← contracts + dispatch + └──────┬──────────┬───────────────────────────┘ + │ │ + ▼ ▼ + ┌──────────┐ ┌──────────────────────────────────┐ + │ store │ │ adapter (AWS Bedrock AgentCore) │ + │ (sqlite) │ │ ↓ │ + └──────────┘ │ cloud worker — runs the task │ + └──────────────────────────────────┘ +``` + +1. **Local assistant** issues `spawn_cloud_agent` over MCP. +2. **MCP server** validates input, hydrates defaults from a runtime profile, calls the core runtime. +3. **Core** picks the right capability + adapter, persists the task, dispatches. +4. **Adapter** invokes the worker in your cloud — synchronous, session, or job target mode. +5. **Worker** runs an agent framework on managed cloud infrastructure. The reference worker in [`worker-agentcore`](https://github.com/agent-dispatch/worker-agentcore) ships with Strands; add support for any other framework by handling its `framework` string. +6. Status, logs, and results flow back through the same path. -Minimal AWS AgentCore session-mode config: +## Configuration + +Minimal AWS Bedrock AgentCore (session mode): ```json { @@ -63,7 +183,7 @@ Minimal AWS AgentCore session-mode config: "backend": "aws-agentcore", "target": { "mode": "session" }, "framework": "strands", - "runtimeTools": { "enabled": ["web-search"] } + "runtimeTools": { "enabled": ["web-search", "code-interpreter"] } } }, "defaults": { @@ -71,3 +191,48 @@ Minimal AWS AgentCore session-mode config: } } ``` + +Once `defaults.runtime` is set, `spawn_cloud_agent` only needs an `instruction`. The `framework` value is passed straight through to the worker — change it to any string your worker recognises. + +Validate the wiring before connecting an MCP client: + +```bash +npx agentdispatch-mcp --config agentdispatch.config.json --check +``` + +## Real-world use cases + +- **Deep research from chat.** Claude Code spawns a cloud agent: "Read the last 90 days of CloudTrail anomalies and propose detection rules." The session keeps going. Twenty minutes later it pulls the result. +- **Parallel codebase audits.** Claude Code fans out a dozen `spawn_cloud_agent` calls, one per service, and aggregates the reports. +- **Long-running OpenClaw runs.** OpenClaw plans the work locally, dispatches the multi-hour execution to a managed AgentCore worker. +- **Tool-rich Hermes jobs.** Hermes uses cloud-side tools (web, code interpreter, repo access) that aren't available to the local process. + +## The rest of AgentDispatch + +This MCP server is one face of a small, focused stack: + +| Repo | Role | +| --- | --- | +| [`core`](https://github.com/agent-dispatch/core) | Runtime contracts and dispatch orchestration | +| [`mcp-server`](https://github.com/agent-dispatch/mcp-server) | **You are here.** MCP face for the runtime | +| [`sdk-js`](https://github.com/agent-dispatch/sdk-js) | TypeScript SDK for application code | +| [`cli`](https://github.com/agent-dispatch/cli) | Command-line interface | +| [`store-sqlite`](https://github.com/agent-dispatch/store-sqlite) | SQLite + filesystem state store | +| [`adapter-aws-agentcore`](https://github.com/agent-dispatch/adapter-aws-agentcore) | AWS Bedrock AgentCore adapter | +| [`worker-agentcore`](https://github.com/agent-dispatch/worker-agentcore) | Standard AgentCore worker contract | +| [`adapter-template`](https://github.com/agent-dispatch/adapter-template) | Starter for new cloud adapters | +| [`docs`](https://github.com/agent-dispatch/docs) | Documentation | + +## Contributing + +PRs, issues, and adapter contributions are welcome. See [`CONTRIBUTING.md`](CONTRIBUTING.md) for the local workflow. + +If you ship a new framework worker (or a new cloud adapter), please open a discussion — we'd love to link it from the org README. + +## License + +Apache-2.0. See [`LICENSE`](LICENSE). + +

+ Built by the AgentDispatch contributors · github.com/agent-dispatch +

diff --git a/brand/README.md b/brand/README.md new file mode 100644 index 0000000..e207af5 --- /dev/null +++ b/brand/README.md @@ -0,0 +1,119 @@ +# AgentDispatch brand kit + +This directory holds the visual identity, the org profile README, and the marketing site for the AgentDispatch organization. Because this branch lives inside `mcp-server` and my publishing access is scoped to a single repo, everything is staged here for you to copy into the right destination repos. + +## Contents + +``` +brand/ +├── logo-mark.svg # square mark for org avatar (dark bg, 200×200, scales) +├── logo-mark-light.svg # square mark for light-bg contexts +├── wordmark.svg # horizontal lockup for light themes (READMEs) +├── wordmark-dark.svg # horizontal lockup for dark themes +├── profile-readme/ +│ └── profile/ +│ ├── README.md # the org profile README +│ └── assets/ # SVGs the README references +└── website/ # static site, no build step + ├── index.html + ├── styles.css + ├── script.js + └── assets/ + ├── favicon.svg + └── og.svg +``` + +The new `mcp-server` README (the viral entry point) is already in place at the repo root and references `brand/wordmark.svg` directly. + +## Design system + +| Token | Value | +| --- | --- | +| Background (dark) | `#0B1020` | +| Surface (card) | `#131A2E` | +| Border | `#1F2A44` | +| Text | `#E2E8F0` | +| Text muted | `#94A3B8` | +| Accent (violet deep / on light) | `#7C3AED` | +| Accent (cyan deep / on light) | `#06B6D4` | +| Accent (violet light / on dark) | `#A78BFA` | +| Accent (cyan light / on dark) | `#22D3EE` | +| Gradient | `#7C3AED → #06B6D4` (or light variants on dark backgrounds) | +| Typeface | Inter (UI / docs), system-ui fallback | +| Code typeface | ui-monospace / JetBrains Mono | + +The logo mark is a **cloud spawning agents**: a soft cloud silhouette (the dispatcher) with three short spokes fanning to three small agent tiles. It reads literally as "cloud → spawn agents" — the system's job in one shape. + +## How to deploy each piece + +### 1. Org avatar + +Upload `brand/logo-mark.svg` (or the light variant) to the AgentDispatch GitHub org settings → *Profile picture*. GitHub will rasterize it for you; the SVG already has a 44px corner radius so it sits well in a rounded display. + +### 2. Org profile README + +GitHub renders `profile/README.md` from a special repo named `.github` at the org root. + +```bash +# Create the .github repo (one-time, manual since this session is scoped to mcp-server) +gh repo create agent-dispatch/.github --public --description "AgentDispatch org profile" + +git clone https://github.com/agent-dispatch/.github +cd .github + +# Copy from this branch +cp -r path/to/mcp-server/brand/profile-readme/profile ./profile + +git add profile +git commit -m "Add org profile README and brand assets" +git push +``` + +Within a minute of pushing, https://github.com/agent-dispatch will render the new profile. + +### 3. Website + +The site is intentionally zero-build — three files plus an assets folder. + +Three easy deployment options: + +**A. GitHub Pages from the docs repo** +```bash +git clone https://github.com/agent-dispatch/docs +cp -r brand/website/* ../docs/ +# enable Pages: Settings → Pages → Source: deploy from a branch → main / root +``` + +**B. A dedicated `website` repo** +```bash +gh repo create agent-dispatch/website --public +cd website +cp -r ../mcp-server/brand/website/* . +git add .; git commit -m "Initial site"; git push +# enable Pages, optionally point a CNAME (agentdispatch.dev / .ai / .io) +``` + +**C. Netlify / Cloudflare Pages / Vercel** +Drag-drop the `brand/website` directory. It's all static — no framework, no build step. + +Recommended: option B with a custom domain. Once live, update README hero links and the OG meta tags in `index.html` to point at the canonical URL. + +### 4. Once everything is live + +In order: + +1. Set the org avatar (step 1). +2. Push the `.github` repo (step 2) — this publishes the wordmark SVGs. +3. Publish the website (step 3) and add the canonical URL to: + - the org profile README ("Learn more →") + - the mcp-server README header + - each repo's `package.json` `homepage` field + +## Approving the look + +If you want to tweak the logo, the two knobs that matter most: + +- **Spoke angles.** Currently 0° middle, ±28° on the right. Tighter (±15°) reads more focused; wider (±45°) reads more "broadcast". Edit the `x2`/`y2` coordinates in `logo-mark.svg`. +- **Palette.** Violet→cyan is on-trend for AI tooling but easy to change — swap the gradient stops in the `` block. The whole brand follows from these two colors. + +That's the entire kit. Ping me if you want a horizontal social-media banner (1500×500 GitHub-org sized), an animated SVG variant, or a dark-mode preview screenshot of the site. diff --git a/brand/logo-mark-light.svg b/brand/logo-mark-light.svg new file mode 100644 index 0000000..9316bf0 --- /dev/null +++ b/brand/logo-mark-light.svg @@ -0,0 +1,50 @@ + + AgentDispatch — spawn cloud agents + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/brand/logo-mark.svg b/brand/logo-mark.svg new file mode 100644 index 0000000..6f5801e --- /dev/null +++ b/brand/logo-mark.svg @@ -0,0 +1,61 @@ + + AgentDispatch — spawn cloud agents + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/brand/profile-readme/profile/README.md b/brand/profile-readme/profile/README.md new file mode 100644 index 0000000..fe86818 --- /dev/null +++ b/brand/profile-readme/profile/README.md @@ -0,0 +1,111 @@ +

+ + + AgentDispatch + +

+ +

Spawn cloud agents from your AI.

+ +

+ Claude Code, OpenClaw, and Hermes plan brilliantly — and choke on multi-hour work.
+ AgentDispatch hands them a managed cloud runtime: one MCP call, results when they land. +

+ +

+ MCP server · + SDK · + CLI · + Core · + Docs +

+ +--- + +## The problem + +Local AI assistants — Claude Code, OpenClaw, Hermes — plan brilliantly. They get cramped the moment work gets long. A deep-research sweep, an account-wide audit, a multi-hour reasoning job: none of it belongs in a laptop's context window. + +## The fix + +AgentDispatch is the one tool your local agent is missing: **spawn a cloud agent with this instruction, come back later for the result.** The run happens on managed cloud compute. Status, logs, and final output stream back over MCP. + +``` +spawn_cloud_agent({ instruction: "Audit every Lambda for public invoke." }) +// → task_id, then status / logs / result on demand +``` + +- ☁️ **Real cloud, not your laptop.** AWS Bedrock AgentCore today; bring your own cloud tomorrow. +- ⏱️ **Built for marathons.** Sync, session, or job mode — same API. +- 🔭 **Full visibility.** Status, logs, results, cancellation. +- 🪶 **Boring defaults.** SQLite, stdio, no hidden services. + +## Made for the agents you already use + +| Local assistant | What it gets | +| --- | --- | +| 🤖 **Claude Code** | A cloud companion for parallel audits, long refactors, and deep-research sweeps. | +| 🦅 **OpenClaw** | Offloads multi-step research and tool-rich runs to a managed cloud worker. | +| 🪽 **Hermes** | Runs long-context reasoning and hours-long jobs without leaving the chat. | +| Claude Desktop · Cursor · Continue · Goose · Zed | Same MCP config, same primitive. | + +## Three faces, one runtime + +- **MCP server** — for AI assistants ([`mcp-server`](https://github.com/agent-dispatch/mcp-server)). +- **TypeScript SDK** — for your own application code ([`sdk-js`](https://github.com/agent-dispatch/sdk-js)). +- **CLI** — for operators and scripts ([`cli`](https://github.com/agent-dispatch/cli)). + +All three call into the same dispatcher and share the same runtime profiles. + +## How it fits together + +``` + Claude Code · OpenClaw · Hermes + │ MCP + ▼ + ┌───────────────────┐ + │ AgentDispatch │ ← spawn / status / logs / result / cancel + └─────────┬─────────┘ + │ + ▼ + ┌───────────────────┐ + │ cloud adapter │ (AWS Bedrock AgentCore today) + └─────────┬─────────┘ + ▼ + ┌───────────────────┐ + │ cloud worker │ runs Strands today · your framework next + └───────────────────┘ +``` + +The adapter contract is small and stable, so a new cloud (or a new framework) is a focused, well-scoped addition — not a rewrite. + +## Repositories + +| Repo | What it is | +| --- | --- | +| [`mcp-server`](https://github.com/agent-dispatch/mcp-server) | **Start here.** MCP face for the runtime — the viral entry point. | +| [`core`](https://github.com/agent-dispatch/core) | Runtime contracts and dispatch orchestration. The heart. | +| [`sdk-js`](https://github.com/agent-dispatch/sdk-js) | TypeScript SDK for application code. | +| [`cli`](https://github.com/agent-dispatch/cli) | Command-line interface for operators. | +| [`store-sqlite`](https://github.com/agent-dispatch/store-sqlite) | Default SQLite + filesystem state store. | +| [`adapter-aws-agentcore`](https://github.com/agent-dispatch/adapter-aws-agentcore) | AWS Bedrock AgentCore adapter. | +| [`worker-agentcore`](https://github.com/agent-dispatch/worker-agentcore) | Standard AgentCore worker contract. Reference Strands worker lives here; add your own framework by handling its `framework` string. | +| [`adapter-template`](https://github.com/agent-dispatch/adapter-template) | Starter for new cloud adapters. | +| [`docs`](https://github.com/agent-dispatch/docs) | Documentation and guides. | + +## Build a new worker or adapter + +A new **framework worker** is a function that maps `{ instruction, context, framework, runtime_tools }` to a result envelope — see [`worker-agentcore`](https://github.com/agent-dispatch/worker-agentcore). + +A new **cloud adapter** implements five methods (dispatch, status, logs, result, cancel) — fork [`adapter-template`](https://github.com/agent-dispatch/adapter-template). + +Open a discussion when you ship — we'd love to link it. + +## Community + +- Issues and PRs welcome on any repo. +- Architecture questions, framework integrations, and ideas → open a discussion in `docs`. + +

+ Apache-2.0 · Built by the AgentDispatch contributors +

diff --git a/brand/profile-readme/profile/assets/logo-mark-light.svg b/brand/profile-readme/profile/assets/logo-mark-light.svg new file mode 100644 index 0000000..9316bf0 --- /dev/null +++ b/brand/profile-readme/profile/assets/logo-mark-light.svg @@ -0,0 +1,50 @@ + + AgentDispatch — spawn cloud agents + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/brand/profile-readme/profile/assets/logo-mark.svg b/brand/profile-readme/profile/assets/logo-mark.svg new file mode 100644 index 0000000..6f5801e --- /dev/null +++ b/brand/profile-readme/profile/assets/logo-mark.svg @@ -0,0 +1,61 @@ + + AgentDispatch — spawn cloud agents + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/brand/profile-readme/profile/assets/wordmark-dark.svg b/brand/profile-readme/profile/assets/wordmark-dark.svg new file mode 100644 index 0000000..ab2f606 --- /dev/null +++ b/brand/profile-readme/profile/assets/wordmark-dark.svg @@ -0,0 +1,63 @@ + + AgentDispatch — cloud agents for AI assistants + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + AgentDispatch + CLOUD AGENTS FOR AI ASSISTANTS + diff --git a/brand/profile-readme/profile/assets/wordmark.svg b/brand/profile-readme/profile/assets/wordmark.svg new file mode 100644 index 0000000..0e5b962 --- /dev/null +++ b/brand/profile-readme/profile/assets/wordmark.svg @@ -0,0 +1,63 @@ + + AgentDispatch — cloud agents for AI assistants + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + AgentDispatch + CLOUD AGENTS FOR AI ASSISTANTS + diff --git a/brand/website/assets/favicon.svg b/brand/website/assets/favicon.svg new file mode 100644 index 0000000..6f5801e --- /dev/null +++ b/brand/website/assets/favicon.svg @@ -0,0 +1,61 @@ + + AgentDispatch — spawn cloud agents + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/brand/website/assets/og.svg b/brand/website/assets/og.svg new file mode 100644 index 0000000..c6e0c42 --- /dev/null +++ b/brand/website/assets/og.svg @@ -0,0 +1,86 @@ + + AgentDispatch — spawn cloud agents from your AI assistant + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + AgentDispatch + + Spawn cloud agents + from your AI. + + Claude Code, OpenClaw, and Hermes plan. AgentDispatch + runs the marathon — on managed cloud compute. + + + + Claude Code + + + OpenClaw + + + Hermes + + + GITHUB.COM/AGENT-DISPATCH + + diff --git a/brand/website/index.html b/brand/website/index.html new file mode 100644 index 0000000..0ab210f --- /dev/null +++ b/brand/website/index.html @@ -0,0 +1,319 @@ + + + + + + AgentDispatch — spawn cloud agents from your AI + + + + + + + + + + + + + +
+ + + AgentDispatch + + +
+ +
+
+ +
+

☁️  CLOUD AGENTS  ·  ONE MCP CALL  ·  BRING YOUR OWN CLOUD

+

+ Spawn cloud agents
+ from your AI. +

+

+ Claude Code, OpenClaw, and Hermes plan brilliantly — and choke on multi-hour work. + AgentDispatch hands them a managed cloud runtime: one MCP call, results when they land. +

+
+ Get started + View MCP server → +
+
+ AWS Bedrock AgentCore + MCP 1.x + Strands today · BYO framework + Apache-2.0 +
+
+
+ +
+

Made for the agents you already ship with

+

+ Plan locally. Run in the cloud. Get results back in chat. No new SDKs, no new auth, no new UI — just one more MCP server in the config you already have. +

+
+
+
Claude Code
+

Fan out parallel cloud agents to audit services, refactor across packages, or grind through long codebase research while the local session stays snappy.

+
+
+
OpenClaw
+

Plan in OpenClaw, execute in the cloud. Deep-research runs and tool-heavy jobs come back as a single tool response — no token bloat in the local session.

+
+
+
Hermes
+

Hermes hands its multi-hour reasoning and tool-rich runs to managed compute — close the laptop, pick the result up later.

+
+
+

Any MCP-compatible client works — these three are the common ones today.

+
+ +
+

The one tool your assistant was missing

+

+ Local agents already have tools. AgentDispatch adds the one they didn't have: a way to start real work in the cloud and pick the result up later. +

+
+
+
☁️
+

Real cloud, not your laptop

+

Managed AWS Bedrock AgentCore today. New clouds plug in through one small adapter contract — your code doesn't move.

+
+
+
⏱️
+

Built for marathons

+

Sessions for stateful runs. Jobs for hours-long work. Sync for fast turn-around. Same API, three modes.

+
+
+
🔭
+

Always observable

+

Status, paginated logs, results, cancellation — over MCP, the SDK, and the CLI. Your assistant always knows what the cloud is doing.

+
+
+
🪶
+

Boring defaults

+

SQLite for state. stdio for transport. Zero hidden services. Runs on a laptop, in CI, and on a server.

+
+
+
+ +
+

Three faces, one runtime

+

Pick the entry point that fits — they all talk to the same dispatcher.

+ +
+ + + +
+ +
+ 
$ npm install @agent-dispatch/core @agent-dispatch/mcp-server \
+    @agent-dispatch/store-sqlite @agent-dispatch/adapter-aws-agentcore
+
+$ npx agentdispatch-mcp --config agentdispatch.config.json --check
+ 
// Claude Code / Claude Desktop / Cursor / any MCP client
+{
+  "mcpServers": {
+    "agentdispatch": {
+      "command": "npx",
+      "args": ["agentdispatch-mcp", "--config", "/abs/path/agentdispatch.config.json"]
+    }
+  }
+}
+
+ +
+ 
$ npm install @agent-dispatch/sdk @agent-dispatch/core
+ 
import { createClient } from "@agent-dispatch/sdk";
+
+const client = createClient({ configPath: "agentdispatch.config.json" });
+
+const { taskId } = await client.spawnCloudAgent({
+  instruction: "Audit every Lambda in this account for public invoke.",
+  runtime: "security-agent"
+});
+
+for await (const event of client.streamTask(taskId)) console.log(event);
+
+ +
+ 
$ npm install -g @agent-dispatch/cli
+
+$ agentdispatch check --config agentdispatch.config.json
+$ agentdispatch spawn "Investigate this stack trace across all services."
+$ agentdispatch logs <task-id> --follow
+$ agentdispatch result <task-id>
+
+
+ +
+

How it fits together

+

+ Your local AI assistant calls a tool. AgentDispatch picks a capability, persists the task, + invokes a worker in the cloud, and streams status and logs back through the same path. +

+
+ + + + + + + + + + + + + + + Claude Code + local assistant + + + OpenClaw + local assistant + + + Hermes + local assistant + + + + + + + + + + AgentDispatch (MCP server + core) + spawn · status · logs · result · cancel + + + + + + + + + + store · SQLite + files + + + adapter · AWS AgentCore + + + cloud worker + runs the long task + + + + + + + +
+
+ +
+

The stack

+

Nine small packages, one focused job each.

+
+ +

mcp-server entry point

+

The MCP face for the runtime. Hand it to Claude Code, OpenClaw, or Hermes.

+ + +

core

+

Runtime contracts and the dispatch loop. The heart of the stack.

+ + +

sdk-js

+

Typed TypeScript SDK for application code.

+ + +

cli

+

Spawn, inspect, and cancel agents from your terminal.

+ + +

store-sqlite

+

Default SQLite + filesystem state store. Zero infrastructure.

+ + +

adapter-aws-agentcore

+

AWS Bedrock AgentCore adapter for the agent-runtime capability.

+ + +

worker-agentcore

+

Standard worker contract — request/response shapes and a small helper.

+ + +

adapter-template

+

Starter for new cloud adapters. Ship one in an afternoon.

+ + +

docs

+

Guides, architecture, and reference material.

+ +
+
+ +
+

Ready to dispatch?

+

Wire AgentDispatch into your assistant and watch the model run work it could never finish on its own.

+
+ Start with the MCP server + Read the docs → +
+
+
+ + + + + + diff --git a/brand/website/script.js b/brand/website/script.js new file mode 100644 index 0000000..caf0e68 --- /dev/null +++ b/brand/website/script.js @@ -0,0 +1,27 @@ +// Tabs +document.querySelectorAll(".tab").forEach((tab) => { + tab.addEventListener("click", () => { + const target = tab.dataset.tab; + document.querySelectorAll(".tab").forEach((t) => t.classList.toggle("active", t === tab)); + document.querySelectorAll(".tab-panel").forEach((p) => { + p.classList.toggle("active", p.dataset.panel === target); + }); + }); +}); + +// Click-to-copy on 
 blocks
+document.querySelectorAll("pre").forEach((pre) => {
+  pre.title = "Click to copy";
+  pre.style.cursor = "copy";
+  pre.addEventListener("click", async () => {
+    const text = pre.innerText;
+    try {
+      await navigator.clipboard.writeText(text);
+      const original = pre.style.borderColor;
+      pre.style.borderColor = "#22D3EE";
+      setTimeout(() => { pre.style.borderColor = original; }, 600);
+    } catch {
+      /* ignore */
+    }
+  });
+});
diff --git a/brand/website/styles.css b/brand/website/styles.css
new file mode 100644
index 0000000..4b0caf7
--- /dev/null
+++ b/brand/website/styles.css
@@ -0,0 +1,322 @@
+:root {
+  --bg: #0B1020;
+  --bg-soft: #11172A;
+  --bg-card: #131A2E;
+  --border: #1F2A44;
+  --text: #E2E8F0;
+  --text-muted: #94A3B8;
+  --text-dim: #64748B;
+  --accent-1: #A78BFA;
+  --accent-2: #22D3EE;
+  --accent-1-deep: #7C3AED;
+  --accent-2-deep: #06B6D4;
+  --maxw: 1120px;
+  --radius: 14px;
+  --radius-lg: 22px;
+}
+
+* { box-sizing: border-box; }
+html { scroll-behavior: smooth; }
+body {
+  margin: 0;
+  font-family: "Inter", ui-sans-serif, system-ui, -apple-system, "Segoe UI", sans-serif;
+  font-feature-settings: "ss01", "cv11";
+  background: var(--bg);
+  color: var(--text);
+  -webkit-font-smoothing: antialiased;
+  -moz-osx-font-smoothing: grayscale;
+  line-height: 1.55;
+}
+
+a { color: inherit; text-decoration: none; }
+code { font-family: ui-monospace, "JetBrains Mono", "SF Mono", Menlo, monospace; font-size: 0.92em; }
+
+/* Nav */
+.nav {
+  position: sticky; top: 0; z-index: 50;
+  display: flex; align-items: center; justify-content: space-between;
+  padding: 18px 28px;
+  background: rgba(11, 16, 32, 0.78);
+  backdrop-filter: saturate(180%) blur(14px);
+  -webkit-backdrop-filter: saturate(180%) blur(14px);
+  border-bottom: 1px solid rgba(255,255,255,0.04);
+}
+.brand { display: flex; align-items: center; gap: 12px; font-weight: 800; letter-spacing: -0.01em; }
+.brand-mark { width: 32px; height: 32px; }
+.brand-text { font-size: 18px; }
+.nav-links { display: flex; gap: 24px; align-items: center; }
+.nav-links a { color: var(--text-muted); font-size: 14px; font-weight: 500; transition: color .15s ease; }
+.nav-links a:hover { color: var(--text); }
+.nav-links .cta {
+  color: #fff;
+  padding: 9px 16px;
+  border-radius: 999px;
+  background: linear-gradient(135deg, var(--accent-1-deep), var(--accent-2-deep));
+  font-weight: 600;
+}
+
+/* Hero */
+.hero {
+  position: relative;
+  padding: 96px 24px 80px;
+  overflow: hidden;
+}
+.hero-inner { max-width: var(--maxw); margin: 0 auto; position: relative; z-index: 1; text-align: center; }
+.eyebrow {
+  display: inline-block;
+  padding: 6px 14px;
+  border-radius: 999px;
+  background: rgba(167, 139, 250, 0.10);
+  border: 1px solid rgba(167, 139, 250, 0.25);
+  color: var(--accent-1);
+  font-size: 13px;
+  font-weight: 600;
+  letter-spacing: 0.02em;
+  margin-bottom: 22px;
+}
+.hero h1 {
+  font-size: clamp(40px, 6vw, 76px);
+  line-height: 1.04;
+  letter-spacing: -0.03em;
+  font-weight: 800;
+  margin: 0 0 22px;
+}
+.grad {
+  background: linear-gradient(135deg, var(--accent-1), var(--accent-2));
+  -webkit-background-clip: text;
+  background-clip: text;
+  color: transparent;
+}
+.lede {
+  font-size: clamp(17px, 1.5vw, 20px);
+  color: var(--text-muted);
+  max-width: 720px;
+  margin: 0 auto 32px;
+}
+.cta-row { display: flex; gap: 14px; justify-content: center; flex-wrap: wrap; }
+.btn {
+  display: inline-block;
+  padding: 13px 22px;
+  border-radius: 12px;
+  font-weight: 600;
+  font-size: 15px;
+  transition: transform .12s ease, box-shadow .15s ease, background .15s ease;
+}
+.btn.primary {
+  background: linear-gradient(135deg, var(--accent-1-deep), var(--accent-2-deep));
+  color: #fff;
+  box-shadow: 0 10px 30px -10px rgba(124, 58, 237, 0.55);
+}
+.btn.primary:hover { transform: translateY(-1px); box-shadow: 0 14px 36px -10px rgba(124, 58, 237, 0.7); }
+.btn.ghost {
+  background: rgba(255,255,255,0.04);
+  border: 1px solid rgba(255,255,255,0.10);
+  color: var(--text);
+}
+.btn.ghost:hover { background: rgba(255,255,255,0.08); }
+.badge-row { display: flex; gap: 10px; justify-content: center; flex-wrap: wrap; margin-top: 36px; }
+.badge {
+  font-size: 12px;
+  font-weight: 600;
+  padding: 6px 12px;
+  border-radius: 999px;
+  background: rgba(255,255,255,0.04);
+  border: 1px solid rgba(255,255,255,0.08);
+  color: var(--text-muted);
+  letter-spacing: 0.02em;
+}
+
+/* Orbit (animated background) */
+.orbit {
+  position: absolute; inset: 0;
+  background:
+    radial-gradient(60% 50% at 50% 0%, rgba(124, 58, 237, 0.25) 0%, transparent 60%),
+    radial-gradient(50% 40% at 80% 30%, rgba(6, 182, 212, 0.18) 0%, transparent 60%),
+    radial-gradient(40% 30% at 15% 35%, rgba(167, 139, 250, 0.18) 0%, transparent 60%);
+  filter: blur(10px);
+  pointer-events: none;
+  animation: float 12s ease-in-out infinite alternate;
+}
+@keyframes float {
+  from { transform: translateY(-6px); }
+  to   { transform: translateY(8px); }
+}
+
+/* Sections */
+section { padding: 72px 24px; }
+.section-title {
+  font-size: clamp(28px, 3.4vw, 40px);
+  letter-spacing: -0.02em;
+  font-weight: 800;
+  margin: 0 0 14px;
+  text-align: center;
+}
+.section-lede {
+  color: var(--text-muted);
+  max-width: 680px;
+  margin: 0 auto 44px;
+  text-align: center;
+  font-size: 17px;
+}
+
+/* Made for */
+.made-for { background: linear-gradient(180deg, transparent, rgba(124,58,237,0.05), transparent); }
+.agent-grid {
+  max-width: var(--maxw);
+  margin: 0 auto;
+  display: grid;
+  gap: 18px;
+  grid-template-columns: repeat(auto-fit, minmax(260px, 1fr));
+}
+.agent-card {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-lg);
+  padding: 24px;
+  transition: transform .15s ease, border-color .15s ease, background .15s ease;
+}
+.agent-card:hover {
+  transform: translateY(-2px);
+  border-color: rgba(167, 139, 250, 0.45);
+}
+.agent-name {
+  font-size: 20px;
+  font-weight: 800;
+  letter-spacing: -0.01em;
+  background: linear-gradient(135deg, var(--accent-1), var(--accent-2));
+  -webkit-background-clip: text;
+  background-clip: text;
+  color: transparent;
+  margin-bottom: 8px;
+}
+.agent-card p { margin: 0; color: var(--text-muted); font-size: 15px; }
+.footnote {
+  text-align: center;
+  color: var(--text-dim);
+  font-size: 14px;
+  margin-top: 28px;
+}
+
+/* Features */
+.features-grid, .repo-grid {
+  max-width: var(--maxw);
+  margin: 0 auto;
+  display: grid;
+  gap: 18px;
+}
+.features-grid { grid-template-columns: repeat(auto-fit, minmax(240px, 1fr)); }
+.repo-grid { grid-template-columns: repeat(auto-fit, minmax(280px, 1fr)); }
+.card, .repo-card {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-lg);
+  padding: 22px 22px 20px;
+  transition: transform .15s ease, border-color .15s ease, background .15s ease;
+}
+.card:hover, .repo-card:hover {
+  transform: translateY(-2px);
+  border-color: rgba(167, 139, 250, 0.45);
+  background: #161E37;
+}
+.card h3, .repo-card h3 {
+  margin: 10px 0 6px;
+  font-size: 18px;
+  letter-spacing: -0.01em;
+  display: flex;
+  align-items: center;
+  gap: 10px;
+}
+.card p, .repo-card p { margin: 0; color: var(--text-muted); font-size: 15px; }
+.card-icon { font-size: 22px; }
+.pill {
+  font-size: 11px;
+  font-weight: 700;
+  padding: 3px 8px;
+  border-radius: 999px;
+  background: linear-gradient(135deg, var(--accent-1-deep), var(--accent-2-deep));
+  color: #fff;
+  letter-spacing: 0.03em;
+}
+.repo-card.primary { border-color: rgba(167, 139, 250, 0.45); }
+
+/* Install / tabs */
+.install { max-width: var(--maxw); margin: 0 auto; }
+.tabs {
+  display: flex;
+  gap: 6px;
+  justify-content: center;
+  margin-bottom: 18px;
+  flex-wrap: wrap;
+}
+.tab {
+  background: transparent;
+  border: 1px solid var(--border);
+  color: var(--text-muted);
+  padding: 9px 16px;
+  border-radius: 999px;
+  font: inherit;
+  font-size: 14px;
+  font-weight: 600;
+  cursor: pointer;
+  transition: color .15s ease, background .15s ease, border-color .15s ease;
+}
+.tab:hover { color: var(--text); }
+.tab.active {
+  color: #fff;
+  background: linear-gradient(135deg, var(--accent-1-deep), var(--accent-2-deep));
+  border-color: transparent;
+}
+.tab-panel { display: none; }
+.tab-panel.active { display: grid; gap: 14px; }
+pre {
+  background: #0A0E1C;
+  border: 1px solid var(--border);
+  border-radius: var(--radius);
+  padding: 18px 20px;
+  overflow-x: auto;
+  margin: 0;
+  font-size: 14px;
+  line-height: 1.55;
+  color: #CBD5E1;
+}
+pre code { white-space: pre; }
+.c-c { color: var(--text-dim); }
+.c-k { color: #A78BFA; }
+.c-s { color: #22D3EE; }
+
+/* Arch */
+.arch { background: linear-gradient(180deg, transparent, rgba(167,139,250,0.04), transparent); }
+.arch-diagram {
+  max-width: var(--maxw);
+  margin: 0 auto;
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-lg);
+  padding: 22px;
+}
+.arch-diagram svg { width: 100%; height: auto; display: block; }
+
+/* CTA section */
+.cta-section { text-align: center; padding-top: 32px; padding-bottom: 96px; }
+.cta-section h2 { font-size: clamp(28px, 3.4vw, 40px); letter-spacing: -0.02em; margin: 0 0 12px; }
+.cta-section p { color: var(--text-muted); margin: 0 0 26px; }
+
+/* Footer */
+.footer {
+  border-top: 1px solid rgba(255,255,255,0.05);
+  padding: 28px 24px;
+  text-align: center;
+  color: var(--text-dim);
+  font-size: 14px;
+}
+.footer p { margin: 4px 0; }
+.footer a { color: var(--text-muted); }
+.footer a:hover { color: var(--text); }
+
+@media (max-width: 640px) {
+  .nav { padding: 14px 18px; }
+  .nav-links { gap: 14px; }
+  .nav-links a:not(.cta) { display: none; }
+  .hero { padding: 60px 18px 50px; }
+  section { padding: 56px 18px; }
+}
diff --git a/brand/wordmark-dark.svg b/brand/wordmark-dark.svg
new file mode 100644
index 0000000..ab2f606
--- /dev/null
+++ b/brand/wordmark-dark.svg
@@ -0,0 +1,63 @@
+
+  AgentDispatch — cloud agents for AI assistants
+  
+    
+      
+      
+      
+    
+    
+      
+      
+    
+    
+      
+      
+    
+    
+      
+      
+    
+  
+
+  
+    
+    
+
+    
+      
+      
+      
+      
+    
+
+    
+      
+      
+      
+    
+
+    
+      
+      
+      
+      
+    
+
+    
+
+    
+      
+      
+      
+      
+        
+        
+        
+      
+    
+  
+
+  AgentDispatch
+  CLOUD AGENTS FOR AI ASSISTANTS
+
diff --git a/brand/wordmark.svg b/brand/wordmark.svg
new file mode 100644
index 0000000..0e5b962
--- /dev/null
+++ b/brand/wordmark.svg
@@ -0,0 +1,63 @@
+
+  AgentDispatch — cloud agents for AI assistants
+  
+    
+      
+      
+      
+    
+    
+      
+      
+    
+    
+      
+      
+    
+    
+      
+      
+    
+  
+
+  
+    
+    
+
+    
+      
+      
+      
+      
+    
+
+    
+      
+      
+      
+    
+
+    
+      
+      
+      
+      
+    
+
+    
+
+    
+      
+      
+      
+      
+        
+        
+        
+      
+    
+  
+
+  AgentDispatch
+  CLOUD AGENTS FOR AI ASSISTANTS
+