Fastmcp Godot Rag

🚀 FastMCP Godot Documentation RAG

FastMCP Godot Documentation RAG is a command-line assistant based on Retrieval Augmented Generation (RAG) for querying and interpreting the official documentation of the Godot engine.

🚀 Quick Start

FastMCP Godot Documentation RAG is a command-line assistant leveraging Retrieval Augmented Generation (RAG) to query and interpret the official documentation of the Godot engine. The project uses Hugging Face semantic vector models and FAISS to build a vector database. Then, through the FastMCP registration tool, it combines the DeepSeek Chat API to return structured answers.

✨ Features

Core Processes

1. Document Preprocessing
   • rst_to_markdown.py batch-converts official .rst documents into structured Markdown, preserving signals, tables, and chapter information.
   • markdown_split.py identifies content types such as chapters, tables, and code in Markdown and splits them into JSON fragments suitable for RAG.
2. Vectorization and Index Building
   • embedding.py reads godot_docs/chunks.json, generates text embeddings using sentence-transformers/all-miniLM-L6-v2, and writes the results to the local FAISS index and metadata files.
3. Retrieval and Answering
   • rag_fastmcp.py loads the local vector library and metadata, registers two FastMCP tools, search_godot_docs and rag_answer, to implement document retrieval and Q&A in combination with the DeepSeek API.

Directory Structure

• godot_docs/: Stores data assets such as converted Markdown, split JSON, and FAISS indexes.
• classes/: A collection of original .rst documents, which can be used as input for data conversion.
• rst_to_markdown.py: A script for batch converting .rst to .md.
• markdown_split.py: A script for intelligent Markdown splitting.
• embedding.py: A script for vectorization and index building.
• rag_fastmcp.py: Loads the index and provides FastMCP tools and a command-line entry.

Running Overview

1. Prepare Data: Place Godot .rst documents in classes/ and execute rst_to_markdown.py to output to godot_docs/md_new/.
2. Split Documents: Run markdown_split.py to generate structured godot_docs/chunks.json.
3. Build Index: Execute embedding.py to generate godot_docs/faiss_db/index.faiss and metadata.pkl.
4. Start the Assistant: Configure DEEPSEEK_API_KEY (or use the built - in demo key in the code) and run rag_fastmcp.py to open the command - line assistant.

After completing the above steps, you can conduct interactive Q&A with the local Godot documentation knowledge base through the FastMCP interface or the command line.