LangGraph

You can host an MCP server that lets you define your own llms.txt sources and execute controlled fetch operations against those sources. This gives you auditable tool calls and clear context when apps query for LLM documentation and guidance.

How to use

You connect MCP to your preferred client (for example Cursor or Windsurf) and run the MCP documentation server locally. The server exposes documentation sources you specify and two primary tool calls you can invoke during a session: list_doc_sources to enumerate available llms.txt sources, and fetch_docs to read the content from those sources. You then reflect on the URLs in llms.txt, consider your input question, and call fetch_docs on any relevant URLs to gather context before forming your answer.

How to install

Prerequisites: you need a runtime environment that provides uv and Node tooling (npm/npx) or Python tooling as applicable to your setup.

1. Install uv if you do not have it yet.

2. Start the MCP document server from the command line to expose your llms.txt sources locally. The server will run and be reachable at http://localhost:8082 by default.

3. (Optional) Run the MCP inspector to test tool calls against the running server.

Additional content

Configuration supports running the server in two modes: an HTTP server that serves at a local URL, and a local stdio-based server that your MCP clients can connect to directly. Use the HTTP mode when you want a standalone server endpoint; use the stdio mode when your client expects a local process to communicate with via standard input/output.

Example setups you can use as starting points include exposing the server at http://localhost:8082 and wiring a local stdio server that runs uvx with the necessary parameters to load the LangGraph llms.txt source.

If you want to test with a specific llms.txt file, you can point the server URL to that file and verify that you can fetch and query its contents through the available tools.

Available tools

list_doc_sources

Lists the available llms.txt sources that MCP will expose to clients, allowing you to audit which documents are included.

fetch_docs

Fetches the content from one or more llms.txt sources so you can review the background material and links before answering questions.