feat: remove create_memory_mcp_toolset wrapper, bump to 0.0.6

CHANGELOG.md

@@ -7,6 +7,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/).
 
 ## [Unreleased]
 
+## [0.0.6] - 2026-05-20
+
+### Breaking
+
+- Removed `create_memory_mcp_toolset()` and all MCP wrapper constants
+  (`ALL_MCP_TOOLS`, `MCP_TOOL_*`) from the package. Use ADK's native
+  `McpToolset` with `SseConnectionParams` pointed at the Agent Memory
+  Server `/sse` endpoint instead. This aligns AMS MCP wiring with the
+  standard pattern used by all other ADK catalog integrations.
+
+### Changed
+
+- Updated `fitness_coach_mcp` example to use native `McpToolset` +
+  `SseConnectionParams` directly.
+
 ## [0.0.5] - 2026-05-19
 
 ### Breaking

README.md

@@ -30,7 +30,6 @@
 | **Sessions** (`RedisWorkingMemorySessionService`) | `BaseSessionService` with auto-summarization and context-window management | Agent Memory Server (REST) |
 | **Long-term memory** (`RedisLongTermMemoryService`) | `BaseMemoryService` with semantic search and recency boosting | Agent Memory Server (REST) |
 | **Memory tools** (`SearchMemoryTool`, `CreateMemoryTool`, ...) | LLM-controlled memory operations | Agent Memory Server (REST) |
-| **AMS MCP toolset** (`create_memory_mcp_toolset`) | Exposes `search_long_term_memory`, `create_long_term_memories`, `edit_long_term_memory`, `delete_long_term_memories`, `get_long_term_memory`, `memory_prompt`, and `set_working_memory` over SSE | Agent Memory Server (MCP) |
 | **RedisVL MCP** (native `McpToolset` against `rvl mcp`) | <ul><li>Tools exposed: `search-records`, `upsert-records` (gate writes with `--read-only`).</li><li>Search modes (one per server, chosen via YAML): `vector` KNN, `fulltext` BM25, or `hybrid` (LINEAR or RRF fusion).</li><li>Server-side query embedding via a configured RedisVL vectorizer; agents never load one locally.</li><li>Schema-aware tool descriptions: filter and return-field hints derived from the bound `IndexSchema`.</li><li>JSON filter language with tag, text, and numeric operators (`eq`, `in`, `between`, `gt`, `lt`, `ne`).</li><li>Transports: stdio, sse, streamable-http. Bearer auth on HTTP. Pagination via `limit` / `offset`.</li></ul> | `rvl mcp` server (`redisvl[mcp]`) |
 | **Search tools with REST** (5 in-process tools) | Vector, hybrid, range, text, SQL search as `BaseTool` subclasses | RedisVL (Python) |
 | **Semantic cache** (`RedisVLCacheProvider`, `LangCacheProvider`) | Skip repeat LLM calls and tool calls by semantic similarity | RedisVL `SemanticCache` or [Redis LangCache](https://redis.io/langcache) |
@@ -287,15 +286,14 @@ For the full decision matrix and runnable demo, see [docs/user_guide/how_to_guid
 
 ## Memory backends
 
-Three ways to ingest, store, and retrieve memory with Agent Memory Server, all interoperable:
+Two ways to ingest, store, and retrieve memory with Agent Memory Server, both interoperable:
 
 | Approach | What it is | Reach for it when |
 |---|---|---|
 | **ADK Services** (`RedisLongTermMemoryService`, `RedisWorkingMemorySessionService`) | The `BaseMemoryService` and `BaseSessionService` implementations. ADK calls AMS for you. | You want framework-managed sessions and automatic memory extraction. Most production cases. |
 | **REST tools** (`MemoryPromptTool`, `SearchMemoryTool`, `CreateMemoryTool`, `UpdateMemoryTool`, `DeleteMemoryTool`, `GetMemoryTool`) | `BaseTool` subclasses that call AMS REST directly. The LLM decides when to invoke them. | You want the agent to control when memory is read or written. |
-| **MCP toolset** (`create_memory_mcp_toolset`) | Same memory operations surfaced over MCP/SSE. Standard MCP tool discovery. | You want one AMS instance shared across many agents, or you prefer MCP wiring over Python tool wrappers. |
 
-The protocol is REST in both of the first two; MCP for the third. All three operate on the same underlying memory.
+Both approaches use REST and operate on the same underlying memory. If you need MCP access to Agent Memory Server, use ADK's native `McpToolset` with `SseConnectionParams` pointed at the AMS `/sse` endpoint.
 
 ---
 
@@ -333,7 +331,7 @@ All examples run via `adk web` and ship with a README and `.env.example`.
 | [`simple_redis_memory`](examples/simple_redis_memory/) | Two-tier memory + auto-summarization |
 | [`travel_agent_memory_hybrid`](examples/travel_agent_memory_hybrid/) | Framework-managed memory: `RedisWorkingMemorySessionService` + `RedisLongTermMemoryService` in a custom FastAPI runner |
 | [`travel_agent_memory_tools`](examples/travel_agent_memory_tools/) | LLM-controlled memory: REST `SearchMemoryTool` / `CreateMemoryTool` / etc. in a default ADK runner |
-| [`fitness_coach_mcp`](examples/fitness_coach_mcp/) | AMS memory exposed over MCP via `create_memory_mcp_toolset` |
+| [`fitness_coach_mcp`](examples/fitness_coach_mcp/) | AMS memory exposed over MCP via ADK's native `McpToolset` |
 | [`redis_search_tools`](examples/redis_search_tools/) | In-process RAG with vector + text + range search tools |
 | [`redis_sql_search`](examples/redis_sql_search/) | `RedisSQLSearchTool` answering catalog questions via parameterized SQL |
 | [`redisvl_mcp_search`](examples/redisvl_mcp_search/) | Same knowledge base as `redis_search_tools/`, served via `rvl mcp` over MCP |

SKILL.md

@@ -29,7 +29,8 @@ links:
   to run [Redis Agent Memory Server](https://github.com/redis/agent-memory-server).
 - The user wants to expose a Redis index to ADK via MCP. For the index
   itself, point ADK's native `McpToolset` at a `rvl mcp` server. For
-  Agent Memory Server's MCP endpoint, use `create_memory_mcp_toolset`.
+  Agent Memory Server's MCP endpoint, use ADK's native `McpToolset`
+  with `SseConnectionParams` pointed at the AMS `/sse` endpoint.
 - The user wants semantic caching for an ADK agent (self-hosted via
   RedisVL or managed via Redis LangCache).
 

docs/for-ais-only/FAILURE_MODES.md

@@ -41,16 +41,15 @@ The `redisvl.extensions.llmcache` path still works but emits a
 to the old path; the regression test in
 `tests/cache/test_provider.py` asserts no `DeprecationWarning` fires.
 
-## One MCP helper, not two
-
-`create_memory_mcp_toolset(...)` exists for Agent Memory Server because
-AMS's MCP URL has a non-trivial `/sse` suffix and the tool-name vocabulary
-is bespoke. There is **no** matching helper for the RedisVL MCP server
-(`rvl mcp`): users wire it with ADK's native `McpToolset` plus
-`StdioConnectionParams` / `SseConnectionParams` /
-`StreamableHTTPConnectionParams`. The maintainers chose this on purpose
-to keep the MCP wiring story aligned with every other ADK catalog
-integration. Do not reintroduce a `create_redisvl_mcp_toolset` wrapper.
+## No MCP wrapper helpers
+
+There are no custom MCP wrapper helpers in this package. For both the
+RedisVL MCP server (`rvl mcp`) and Agent Memory Server's MCP endpoint,
+users wire ADK's native `McpToolset` with `StdioConnectionParams` /
+`SseConnectionParams` / `StreamableHTTPConnectionParams` directly. This
+keeps the MCP wiring story aligned with every other ADK catalog
+integration. Do not introduce `create_redisvl_mcp_toolset` or
+`create_memory_mcp_toolset` wrappers.
 
 ## Two cache providers exist on purpose
 

docs/for-ais-only/REPOSITORY_MAP.md

@@ -33,8 +33,7 @@ src/adk_redis/
                           service (MemoryPromptTool, SearchMemoryTool,
                           CreateMemoryTool, DeleteMemoryTool,
                           UpdateMemoryTool).
-    mcp_memory.py         MCP tool surface for the same memory operations
-                          (Agent Memory Server).
+
   cache/
     __init__.py           Re-exports the cache providers.
     _provider.py          Provider protocol and base class.
@@ -81,8 +80,7 @@ tests/
 | Working-memory session storage | `sessions/working_memory.py` |
 | Long-term memory + Memory Server proxy | `memory/long_term_memory.py`, `memory/_utils.py` |
 | ADK Memory tools (FunctionTool wrappers) | `tools/memory/` |
-| MCP memory tool surface (Agent Memory Server) | `tools/mcp_memory.py` |
-| MCP search (RedisVL `rvl mcp` server) | Use ADK's native `McpToolset` directly; no adk-redis wrapper. |
+| MCP (RedisVL or AMS) | Use ADK's native `McpToolset` directly; no adk-redis wrapper. |
 | Vector / Hybrid / Range / Text / SQL search tools | `tools/search/` |
 | LLM and tool semantic caching | `cache/llm_cache.py`, `cache/tool_cache.py` |
 | Cache provider abstraction (RedisVL / LangCache) | `cache/_provider.py` |

docs/user_guide/01_integration.md

@@ -293,13 +293,15 @@ agent = Agent(
 
 ### 3. MCP-Based Tools (Protocol-Based)
 
-Best for: MCP ecosystem integration and standardized tool discovery.
+Best for: MCP ecosystem integration and standardized tool discovery. Use ADK's native `McpToolset` with `SseConnectionParams` to connect to Agent Memory Server's SSE endpoint.
 
 ```python
-from adk_redis import create_memory_mcp_toolset
+from google.adk import Agent
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
 
-memory_tools = create_memory_mcp_toolset(
-    server_url="http://localhost:8000",
+memory_tools = McpToolset(
+    connection_params=SseConnectionParams(url="http://localhost:8000/sse"),
     tool_filter=["search_long_term_memory", "create_long_term_memories"],
 )
 
@@ -317,7 +319,7 @@ See the [fitness_coach_mcp example](../examples/fitness_coach_mcp/) for a comple
 |----------|---------------------|
 | Full ADK integration | Memory Services |
 | LLM decides when to remember | REST Tools |
-| MCP ecosystem | MCP Tools |
+| MCP ecosystem | MCP Tools (native `McpToolset`) |
 | Debugging/development | REST Tools |
 | Multi-agent systems | MCP Tools |
 
@@ -360,11 +362,12 @@ MCP provides a standardized protocol for connecting agents to tools via Server-S
 
 ```python
 from google.adk import Agent
-from adk_redis import create_memory_mcp_toolset
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
 
 # Connect to Agent Memory Server's MCP endpoint
-memory_tools = create_memory_mcp_toolset(
-    server_url="http://localhost:9000",  # MCP server port
+memory_tools = McpToolset(
+    connection_params=SseConnectionParams(url="http://localhost:9000/sse"),
     tool_filter=["search_long_term_memory", "create_long_term_memories"],
 )
 

examples/fitness_coach_mcp/README.md

@@ -52,13 +52,15 @@ Open **http://localhost:8000** in your browser.
 
 ## How MCP Memory Works
 
-This example uses `create_memory_mcp_toolset()` to connect to the Agent Memory Server's MCP endpoint:
+This example uses ADK's native `McpToolset` to connect to the Agent Memory Server's MCP endpoint:
 
 ```python
-from adk_redis.tools.mcp_memory import create_memory_mcp_toolset
+from google.adk import Agent
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
 
-memory_tools = create_memory_mcp_toolset(
-    server_url="http://localhost:8088",
+memory_tools = McpToolset(
+    connection_params=SseConnectionParams(url="http://localhost:8088/sse"),
     tool_filter=[
         "search_long_term_memory",
         "create_long_term_memories",

examples/fitness_coach_mcp/fitness_coach/agent.py

@@ -15,7 +15,7 @@
 """Personal Fitness Coach Agent with MCP Memory Tools.
 
 This agent demonstrates MCP-based memory integration:
-- Uses create_memory_mcp_toolset() for memory operations
+- Uses ADK's native McpToolset to connect to Agent Memory Server
 - Stores semantic memories (profile, injuries, equipment, goals)
 - Stores episodic memories (workouts with event_date, milestones)
 - Searches memory before making recommendations
@@ -24,17 +24,18 @@
 import os
 
 from google.adk import Agent
-
-from adk_redis.tools.mcp_memory import create_memory_mcp_toolset
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
 
 # Configuration from environment
 # MCP server runs on port 9000 (separate from REST API on 8088)
 MEMORY_MCP_URL = os.getenv("MEMORY_MCP_URL", "http://localhost:9000")
 
 # Create MCP toolset for Agent Memory Server
 # Connects to the MCP server's SSE endpoint
-memory_tools = create_memory_mcp_toolset(
-    server_url=MEMORY_MCP_URL,
+sse_url = MEMORY_MCP_URL.rstrip("/")
+memory_tools = McpToolset(
+    connection_params=SseConnectionParams(url=f"{sse_url}/sse"),
     tool_filter=[
         "search_long_term_memory",
         "create_long_term_memories",

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "adk-redis"
-version = "0.0.5"
+version = "0.0.6"
 description = "Redis integrations for Google's Agent Development Kit (ADK)"
 readme = "README.md"
 license = "Apache-2.0"