MCP/Doc/How To/Set up with mise and Claude Code with Langgraph llms.txt

• How To Serve LangGraph & LangChain llms-txt via MCP with mise and validate in Claude Code

  • Goal

    • Stand up a local MCP documentation server using a YAML config for both LangGraph Python and LangChain Python llms.txt, wire it into Claude Code through mise, and confirm the connection with MCP Inspector.

  • Preconditions

    • You have mise ≥ 2024.6 installed and configured.
    • Your mise.toml already defines a claude task like:
      • toml

        [tasks.claude]
        # run mise install && mise reshim before running this.
        description = "Run Claude Code CLI"
        run = "npx @anthropic-ai/claude-code $@"
        [tasks.claude.env]
        CLAUDE_CODE_USE_BEDROCK    = "true"
        AWS_PROFILE                = "bedrock"
        AWS_REGION                 = "us-east-1"
        ANTHROPIC_MODEL            = "us.anthropic.claude-3-7-sonnet-20250219-v1:0"
        ANTHROPIC_SMALL_FAST_MODEL = "us.anthropic.claude-3-5-haiku-20241022-v1:0"

    • You have Node.js ≥ 18 and uvx available (via mise's node plugin).
    • This guide assumes you already have the uv and python tools installed and shimmed via your home directory mise.toml, so that uvx is available on your PATH.
      • See: mise Getting Started – Adding tools

  • Procedure

    • 1. Add a YAML configuration for documentation sources

      • Create a docs.yaml file in the project root:
        • yaml

          # MCP documentation sources
          - name: LangGraph Python
            llms_txt: https://langchain-ai.github.io/langgraph/llms.txt
          - name: LangChain Python
            llms_txt: https://python.langchain.com/llms.txt

    • 2. Extend mise.toml with MCP tasks

      • Add the following to your mise.toml:
        • toml

          [tasks.mpc-docs]
          description = "Serve LangGraph & LangChain llms.txt via MCP"
          alias       = "mcp"
          run = '''
          uvx --from mcpdoc mcpdoc \
            --yaml docs.yaml \
            --transport sse \
            --port ${MPCDOC_PORT:-8082} --host localhost
          '''
          [tasks.mpc-inspect]
          description = "Launch MCP Inspector UI"
          run = "npx -y @modelcontextprotocol/inspector mise run mpc-docs"

    • 3. (Optional) Add a local environment file

      • If you want to set a custom port or other environment variables, create a .env file (or reuse an existing one) with:
        • env

          # optional override
          MPCDOC_PORT=8082

      • Note: By default, mise does NOT automatically load .env files. To have mise load this file for your tasks, add the following to your mise.toml:
        • toml

          # This is forgiving: if .env does not exist, mise will ignore it without error.
          # See: https://mise.jdx.dev/configuration/environments.html#setting-envvars
          [env]
          _.file = ".env"

    • 4. Other ways to customize the port for the mpc-docs task

      • You can set the MPCDOC_PORT environment variable for the mpc-docs task in several mise-specific ways:
        • Directly in the task definition:
          • toml

            [tasks.mpc-docs.env]
            MPCDOC_PORT = "8083" # or any available port

        • Via the command line for a one-off run:
          • sh

            MPCDOC_PORT=8084 mise run mpc-docs

        • With a .env file as described above.
      • See mise: Setting EnvVars for more options.

    • 5. Start the MCP server

      • Run:
        • sh

          mise run mpc-docs

      • You should see Uvicorn log output showing that it's serving on your selected port and has loaded two documentation sources.

    • 6. Inspect and validate

      • In another terminal, run:
        • sh

          mise run mpc-inspect

      • MCP Inspector opens in your browser at http://127.0.0.1:6274. Click Connect, then test list_doc_sources and fetch_docs to verify both sources load correctly.

    • 7. Connect Claude Code to the project server

      • 1. Add the MCP server to your project's .mcp.json using:
        • sh

          claude mcp add-json langgraph-docs \
            '{"type":"sse","url":"http://localhost:8082"}' \
            -s project

      • See: Claude Code Tutorials – Set up Model Context Protocol MCP (project scope)
      • 2. Create a CLAUDE.md file in the root of your repo with usage rules like:
        • markdown

          for ANY question about LangGraph or LangChain, use the langgraph-docs server to help answer –  
          + call list_doc_sources to view available llms.txt files  
          + call fetch_docs to read them  
          + reflect on the input question and fetched content before answering

      • 3. Run:
        • sh

          mise run claude

      • 4. In Claude's terminal, run /mcp status and confirm langgraph-docs is connected. Ask:

        • "What types of memory does LangGraph support?"
          Claude should respond using the local docs without external fetches.
          

  • Troubleshooting

    • Port already in use — You can resolve port conflicts by:
      • Setting a different value for MPCDOC_PORT in your .env file (and ensuring it's loaded as above)
      • Setting MPCDOC_PORT in the [tasks.mpc-docs.env] section of your mise.toml
      • Passing it inline for a single run: MPCDOC_PORT=8085 mise run mpc-docs
    • Inspector can't connect — verify mpc-docs is running and using the expected port.
    • Claude doesn't see the server — confirm .mcp.json includes the correct langgraph-docs entry and matches the transport/URL.

  • References

    • mcpdoc GitHub
    • llms.txt spec
    • LangGraph llms.txt
    • LangChain llms.txt