MCP Claude Managed Agents

An MCP server that exposes the Claude Managed Agents documentation to Claude Desktop and Claude Code, keeping your AI assistant always up to date with the latest API docs.

What it does

Provides three tools that Claude can call to read current documentation directly from platform.claude.com:

Tool	Description
list_claude_sections	List all available Managed Agents documentation sections
get_claude_page	Fetch the full content of a documentation page by slug or URL
search_claude_docs	Search sections by keyword and get matching slugs

Covered sections (20 pages): overview, quickstart, onboarding, agent-setup, tools, mcp-connector, permission-policies, skills, environments, cloud-containers, sessions, events-and-streaming, define-outcomes, vaults, github, files, memory, multi-agent, migration, observability.

The catalog is seeded at startup with all known sections and auto-discovers new pages from the live platform.claude.com/sitemap.xml.

Requirements

• Python 3.11+
• uv (recommended) or pip

Installation

git clone https://github.com/Attilio81/MCP_CMA.git
cd MCP_CMA
uv pip install -e .
# or
pip install -e .

This installs the mcp-claude command.

Claude Desktop configuration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "claude-docs": {
      "command": "mcp-claude"
    }
  }
}

If installed in a virtual environment, use the full path:

{
  "mcpServers": {
    "claude-docs": {
      "command": "/path/to/venv/bin/mcp-claude"
    }
  }
}

Restart Claude Desktop — the three tools appear automatically.

Claude Code configuration

claude mcp add claude-docs mcp-claude

Or add to your project's .mcp.json:

{
  "mcpServers": {
    "claude-docs": {
      "command": "mcp-claude"
    }
  }
}

How it works

1. At startup, seed entries for all 20 known Managed Agents pages are loaded into an in-memory catalog.
2. The public sitemap at platform.claude.com/sitemap.xml is fetched to discover any new pages added by Anthropic. If the sitemap is unavailable, the server starts with seed entries only (no crash).
3. On each tool call, pages are fetched live from platform.claude.com via plain HTTP (Next.js SSR — no JavaScript rendering required).
4. Fetched content is cached in memory for the duration of the server session. Restart the server to clear the cache.

Running tests

pip install pytest
pytest

License

MIT