mcp

The Strands Agents MCP Server provides curated documentation access to your GenAI tools by indexing llms.txt content and exposing it through the MCP protocol. It uses a smart document search with TF-IDF scoring that prioritizes titles, headers, and code blocks, and it supports on-demand fetching of full document content to keep responses snappy. You can query the server for relevant tool docs, generate contextual snippets, and retrieve real URLs when needed. Typical use involves configuring an MCP client to point at this server (via uvx or another MCP client) and then using the client's search_docs and fetch_doc capabilities to locate and retrieve documentation for Strands Agents tools and their APIs. The server is designed to work with a variety of MCP clients and provides auto-approval of common actions to streamline workflows.

Tools and capabilities available include:

• search_docs: perform a structured search over indexed llms.txt content with relevance ranking.
• fetch_doc: retrieve the full document content for a selected result, with lazy-loading to optimize performance.
• Snippet generation: receive concise, context-rich snippets for quick comprehension of the documentation.
• Real URL support: access to documentation hosted at real HTTPS URLs while still indexing locally for fast search. To use it, configure your MCP client to start the server using the uvx command with the strands-agents-mcp-server package, then issue search_docs queries and retrieve results with fetch_doc as needed.

Prerequisites:

• Install uv following the official instructions (https://github.com/astral-sh/uv).
• Ensure you have a compatible Python environment if you prefer running the Python module directly (optional for some setups).

Installation steps (recommended path using uvx):

1. Install the Strands Agents MCP Server package (Python package available on PyPI):
   • Run: pip install strands-agents-mcp-server
   • This provides the strands-agents-mcp-server entry point for uvx.
2. Verify installation by listing installed packages or invoking the server helper:
   • Run: uvx strands-agents-mcp-server --help
3. Start the MCP server with uvx:
   • Run: uvx strands-agents-mcp-server
   • Optional: configure environment variables such as FASTMCP_LOG_LEVEL for logging verbosity.

Alternative local development (Python module):

1. Install the package as above.
2. Run the server module directly (if supported by your setup):
   • Run: python -m strands_mcp_server

If you prefer using a package manager, you can also test with the MCP Inspector as described in the Quick Testing section.

Notes and tips:

• The server expects uv as a prerequisite; ensure it is installed and accessible in your PATH.
• If you run into permission or environment issues, set FASTMCP_LOG_LEVEL (e.g., INFO or DEBUG) to aid troubleshooting.
• Some MCP clients may require you to register the mcpServers entry under a user profile or config file; consult your client’s documentation for exact steps.
• The server is designed to be beginner-friendly with autoApprove hooks for search_docs and fetch_doc to streamline common workflows.
• For testing, you can use the MCP Inspector to validate connectivity and protocol exchanges with commands like npx @modelcontextprotocol/inspector uvx strands-agents-mcp-server or python -m strands_mcp_server for local development.