Rationale

I want my LLM to be able to read, modify and update my code. this implies it should be able to:

• read and write files and directories
• run code (at least python)
• install dependencies (uv or venv)

Now on how I want to do it: This should run on its own : no need for an additional machine or API access. The LLM must not have the ability to run containers. The setup must itself be running inside a container with a mounting point for data to easily run on a home server.

Organisation

2 sets of tools

• shell prompt
• code execution

Shell prompt (WIP)

Just provide a shell prompt with the ability to set the current active directory.

Python execution (WIP)

No complex security, the python code must run and the LLM must be able to add its own deps to run the code.

Integration with Librechat

To integrate MCP Python with Librechat, you need to update the following configuration files:

docker-compose.yml

Add the following service configuration to your docker-compose.yml:

  mcp_python:
    container_name: mcp_python
    image: docker.nautil.org/mcp-python:latest
    environment:
      - WORKDIR=/home/projects
    volumes:
      - ./projects:/home/projects

librechat.yaml

Add the following MCP server configuration to your librechat.yaml:

mcpServers:
  python:
    type: streamable-http
    url: http://mcp_python:8000/mcp

This setup allows Librechat to interact with the MCP Python service, providing code execution and file management capabilities to your LLM.