Ucp-Local – Offline RAG for Claude Desktop, Cursor, and LM Studio (github.com)

UCP-Local has been launched as an offline retrieval-augmented generation (RAG) solution designed for integration with local AI models and clients like Claude Desktop, Cursor, and LM Studio. This local-first MCP server indexes user files—such as notes, code, and conversation history—allowing for seamless retrieval and grounding of language model responses. It utilizes hybrid retrieval methods combining BM25 and vector search, ensuring efficient indexing and search capabilities without any reliance on cloud services, thereby addressing privacy concerns in sensitive industries. The significance of UCP-Local lies in its fully offline functionality, which allows users to utilize local models while maintaining complete control over their data. It supports features like conversation memory, which allows for the searching of previous AI chats across future sessions, and efficient tree-sitter-aware code chunking. Additionally, the compact installation process and the absence of telemetry or cloud dependencies make it ideal for environments with strict privacy regulations, such as legal or medical workflows. Overall, UCP-Local represents a significant step towards empowering users with local, secure AI capabilities without compromising on performance or accessibility.