Codebase RAG MCP Server

A Retrieval-Augmented Generation (RAG) Model-Controller-Provider (MCP) server designed to assist AI agents and developers in understanding and navigating codebases.

Quick Links

• Getting Started: Installation and Basic Usage.

• User Manual: Tutorials, Features, and Advanced Usage.

• Configuration: Environment variables and performance tuning.

• Architecture: System design and high-level overview.

• Contributing: Developer setup and guidelines.

Related MCP server: PAMPA

Overview

This tool allows you to "chat" with your code. It indexes your local directories or GitHub repositories into a vector database (Qdrant), allowing Large Language Models to perform semantic searches and retrieve accurate code snippets.

Key Features

• 🔍 Semantic Search: Find code by meaning, not just keywords.

• 🎯 Two-Stage RAG: 22-31% accuracy improvement with cross-encoder reranking (Qwen3-Reranker).

• 🧠 Intelligent Chunking: Parses code into functions and classes using Tree-sitter for better context.

• ⚡ High Performance: Supports incremental indexing and MPS acceleration on macOS.

• 🌐 Multi-Language: Supports Python, JavaScript, TypeScript, Go, Rust, Java, C++, and more.

• 📝 File Logging: Optional centralized logging with automatic rotation for debugging.

Quick Start

1. Setup:

   ./setup.sh

2. Index Code:

   uv run python manual_indexing.py -d "." -m clear_existing

3. Connect: Follow the instructions from setup.sh to add the server to your MCP client (e.g., Claude Desktop).

For detailed instructions, please visit Getting Started.

Documentation

Detailed documentation is available in the docs/ directory:

• MCP Tools Reference

• Architecture Deep Dive

License

This project is licensed under the MIT License - see the LICENSE file for details.