FastMCP can expose search and document-reading functions inside a Python backend. Qdrant's official MCP server is a memory-oriented store/find example, not a complete repository reader; its read-only mode does not add source navigation or custom hybrid retrieval.

Keep the MCP layer thin

FastMCP supports mounting into an existing Python web application. I propose using it only as the transport wrapper around the application's own search and revision-aware reading functions, without changing publication.

Qdrant's official MCP server documents qdrant-store and qdrant-find, configurable embeddings through FastEmbed, and QDRANT_READ_ONLY to disable writes. It is useful reference code, but its documented tool surface is not a complete Markdown repository search-and-read service.

Keep model selection and query encoding aligned with ingestion. Expose source-reading and metadata filtering deliberately rather than assuming any MCP wrapper around the same database reproduces the intended retrieval behavior.