Home / MCP / LangGraph Docs MCP Server
Provides access to llms.txt sources and fetch_docs tool to securely read and audit documentation contexts.
Configuration
View docs{
"mcpServers": {
"langgraph_docs_mcp": {
"command": "uvx",
"args": [
"--from",
"mcpdoc",
"mcpdoc",
"--urls",
"LangGraph:https://langchain-ai.github.io/langgraph/llms.txt LangChain:https://python.langchain.com/llms.txt",
"--transport",
"stdio"
]
}
}
}You can run a dedicated MCP server to manage and audit the tools that fetch and read llms.txt documentation sources. This gives you full control over where your documentation comes from and how tool calls are executed, while enabling you to audit the context returned for each request.
You run an MCP server locally and connect your client (such as Cursor, Windsurf, or Claude Desktop) to it. Your client can then request documentation from the specified llms.txt sources and invoke tools like list_doc_sources and fetch_docs to read and reflect on the URLs inside those files. Use the provided MCP inspector to test the server and verify tool calls before connecting your client.
Prerequisites: ensure you have a runtime environment that can execute the MCP tooling used here. You will typically need a Node-based environment and the UVX launcher.
1) Install the UVX launcher if you do not already have it.
2) Start the MCP server with the standard command that wires in the llms.txt sources. The example below uses the langgraph and langchain sources and runs the server in stdio mode.
uvx --from mcpdoc mcpdoc \
--urls "LangGraph:https://langchain-ai.github.io/langgraph/llms.txt LangChain:https://python.langchain.com/llms.txt" \
--transport stdioConfiguration and usage notes concentrate on controlling which domains you can access and how you invoke the tools. You can explicitly allow domains, or use a wildcard to permit all domains if you accept the security risk.
Security and domain access controls ensure that only approved llms.txt sources are queried. When using remote llms.txt URLs, only the domain of the provided URL is allowed by default. For local files, you must specify allowed domains yourself via the MCP configuration.
You can test the MCP server locally with the inspector to verify the available tools and responses before connecting your client.
The server exposes two core tools for working with llms.txt sources and their contents.
- Use list_doc_sources to discover which llms.txt files are available.
- Use fetch_docs to read and retrieve documentation from the URLs listed in an llms.txt file.
- Reflect on the URLs, analyze your input question, and then fetch relevant documentation to inform your answer.
Tool that lists all available llms.txt sources the MCP server can use, allowing you to audit and choose the sources you trust.
Tool that reads documents from the URLs provided by the selected llms.txt sources, returning the documentation content for analysis.