CodeRAG

CodeRAG is an open-source code search engine that indexes repositories using AST (abstract syntax tree) chunking, supporting more than 15 programming languages. It combines traditional keyword-based TF-IDF with dense vector similarity to deliver <50ms hybrid search results. Through its Model Context Protocol (MCP) interface, AI coding assistants can directly query the index, read specific file ranges, and search documentation—making it a ready-to-use RAG backend for developer tools.

Key strengths include zero external dependencies, local execution, and adaptive snippet sizing. The MCP layer exposes seven structured tools (e.g., search_code, read_code, search_docs) so LLMs can fetch exactly what they need, with results dynamically extended or truncated based on ranking and with options to exclude test directories or filter by language.

Limitations: repositories must be pre-indexed, which can be time- and storage-intensive for large codebases. The project is at an early stage (11 GitHub stars), so community support and long-term stability are not guaranteed. Users must self-host and maintain the service.

CodeRAG is free, open-source, and runs locally with no cloud costs.