RAG Server
0
Ingest source code and documentation to build a local vector index, exposing semantic search and file access to LLMs via Model Context Protocol (MCP) tools.
About
This lightweight, zero-network RAG server enables Retrieval-Augmented Generation for any repository. It processes source code and documentation, chunks the content, builds local embeddings, and provides MCP tools for semantic search, secure file reading, and directory listing. Designed to integrate seamlessly with MCP-aware clients like GitHub Copilot Agent mode, it enhances LLM capabilities with accurate, context-aware information directly from your codebase.
Key Features
- Pure local embedding inference (no external API calls)
- Multi-language source + docs support with configurable exclusions
- Fast glob file discovery and overlapping chunking for better recall
- Optional persistent JSON index with warm start and incremental reindexing
- Stdio or Streamable HTTP transport with robust path handling and security features
- 0 GitHub stars
Use Cases
- Integrating local codebases and documentation with LLM agents (e.g., GitHub Copilot)
- Enabling semantic search and contextual file access for custom AI-powered IDEs or tooling
- Providing a secure and efficient knowledge base for LLMs to answer questions about a repository