QMD is a local-first search engine created by Tobi Lütke, CEO of Shopify, designed to make every document on your machine instantly findable through a hybrid retrieval pipeline. It combines three search strategies in a single query: BM25 keyword matching for precise term lookups, vector semantic search for meaning-based retrieval, and LLM-powered re-ranking that understands context and intent. All processing runs entirely on-device using node-llama-cpp with GGUF quantized models, requiring no cloud services, API keys, or network connectivity. The engine indexes markdown files, meeting transcripts, knowledge bases, and any text-based documentation you point it at.
What sets QMD apart from conventional search tools is its first-class MCP server integration. Claude Code, Cursor, Windsurf, and any MCP-compatible AI editor can query your indexed documents directly through the Model Context Protocol, giving coding assistants access to your personal knowledge base without uploading anything to external servers. The MCP server supports both stdio transport for single-client use and HTTP transport at localhost:8181/mcp for concurrent multi-client sessions. An optional intent parameter helps disambiguate ambiguous queries, available through the CLI with --intent flag, MCP tool fields, and the programmatic API.
Built with TypeScript and requiring Node.js 22+, QMD maintains an active development pace with regular releases adding features like improved chunking strategies and expanded model support. The project has attracted over 19,000 GitHub stars since its release, reflecting strong developer interest in privacy-preserving search solutions. It ships under the MIT license with zero telemetry, making it suitable for sensitive corporate environments where document contents cannot leave local infrastructure. The CLI interface supports batch indexing, incremental updates, and configurable search pipelines that can be tuned per collection.