Godot RAG

You run a dedicated MCP server that serves Godot documentation to the Godot RAG model. This server makes it easier for the model to access up-to-date Godot guidance during interactions and improves answer quality when you query Godot-related topics.

How to use

Start or connect to the server from your MCP client to access Godot documentation through the MCP workflow. Use the server to query, retrieve, and summarize Godot docs relevant to your questions. The server is designed to be run locally and to read from a prepared vector store so the model can retrieve precise information efficiently. You will typically start the server with its Python entry point and point it at your local Chroma vector store and collection.

How to install

Prerequisites: you need Python available on your system. The recommended workflow uses a virtual environment tool to isolate dependencies.

uv venv --python 3.12
source ./.venv/bin/activate
uv sync
cp .env.example .env.local

Configuration and startup guidance

Configure the MCP server to point to your local vector store and collection. The server exposes a standard startup pattern where you pass the path to the vector store and the collection name.

{
  "mcpServers": {
    "godot_rag": {
      "command": "python",
      "args": [
        "<path to the server script 'main.py'>",
        "-d",
        "<path to the chroma_db on your computer>",
        "-c",
        "<name of the collection in the chroma_db>"
      ]
    }
  }
}

Start or run the MCP server

Once dependencies are installed and the environment is prepared, start the server with the following example command. This launches the server and loads the specified vector store and collection so you can begin querying Godot documentation.

# start mcp server
python main.py -d artifacts/vector_stores/chroma_db -c artifacts_chunks_SZ_400_O_20_all-MiniLM-L6-v2

Notes on usage patterns

• You can replace the vector store path and collection name with your own prepared store when you bring up the server in your environment.

• Use the MCP client to issue queries that leverage the Godot documentation corpus to guide responses from the Godot RAG model.

Additional configuration options

The server configuration snippet shows how to declare a local, self-contained MCP server. You can adapt the values to point to your own document corpus and collection.

Security and best practices

Keep your vector store and collection names secure and accessible only within trusted environments. If you expose the server externally, ensure you have appropriate authentication and access controls in place.

Troubleshooting tips

If the server fails to start, verify that the Python environment is active, the path to main.py exists, and the vector store and collection paths are correct. Check for missing dependencies and ensure the specified collection name matches the one in your chroma_db.

Examples of supported models

All-MiniLM-L6-v2 is listed as a compatible vectorization model option, with other options such as bge-m3 and bge-large-zh-v1.5 available as alternatives.

Configuration for development and testing

During development and testing, you may use a local inspector to validate the MCP setup and inspect the server workflow.