rag-mcp

An MCP server that retrieves Godot engine documentation from a ChromaDB database. Setup means preparing a vector store with web2embeddings and registering the server in VS Code Insiders.

Godot RAG MCP is a Model Context Protocol tool that pulls relevant Godot information from a local ChromaDB database. The assistant queries the vector store and gets matching documentation content back as tool results, which is the retrieval half of a RAG setup. The tool is written for Godot, but the README notes it can be pointed at any chroma database with small changes to the tool description, so the retrieval setup is reusable. Because the database is local, the whole loop runs without sending the Godot corpus to a remote service. The result is documentation lookup that behaves like a tool call.

Three steps get it running. Prepare the vector store with the web2embeddings tool, clone or download the repository onto the local disk, and register the MCP server in Visual Studio Code Insiders. The vector store has to exist before the server can retrieve from it, so the embedding step comes first. There is no build step to worry about, since the tool is Python and the heavy lifting happens in the vector store.

The configuration targets VS Code Insiders specifically. The README notes the Insiders build may need to be downloaded separately to get MCP or agent mode, which scopes the tool to an editor where the server can actually be registered and used by the assistant at runtime. Anyone who has registered an MCP server before will recognize the pattern; the Insiders requirement is the one quirk to plan around.

• Source 1 · Project repository
• Source 2 · README