Rag Hub MCPSelf-hosted RAG that speaks MCP

Drop folders on disk, get a searchable knowledge base served to any MCP-compatible agent or over REST. Zero infrastructure.

Folders are knowledge bases

1st-level folder = 1 KB, named after the folder. Drop documents, let indexing pick them up.

9 tools over the Model Context Protocol. Works with opencode, Claude Code, Cline and more.

Vector cosine similarity fused with SQLite FTS5 keyword search, weighted and scored per chunk.

Markdown, text, HTML, code, PDF, DOCX, XLSX, PPTX — extracted to plain text.

Your documents never leave your machine. Embeddings come from an OpenAI-compatible endpoint you control.

One SQLite database with FTS5. No vector database, no separate services, no server to keep running.