You index scattered learning materials into one RAG corpus, then query from Cursor, VS Code, or any MCP client. It handles PDFs, GitHub repos, YouTube videos (with optional vision that merges on-screen code and diagrams into transcript chunks), Discord exports, and web docs. Built on LangChain and Chroma with local Nomic embeddings. Five tools cover add, query, list, remove, and tag operations. Query filters by document ID, tag, type, or PDF page range. Citations point to page numbers, timestamps, file names, or URLs. Runs via uvx with OpenRouter by default, but you can swap in OpenAI, Anthropic, or Cerebras. Supports FlashRank reranking, multi-query expansion, and LangSmith tracing through env vars.
claude mcp add --transport stdio ndjordjevic-pinrag uvx pinrag