A self-hosted MCP server for PageIndex's vectorless, reasoning-based document retrieval. It ingests PDFs into a hierarchical table of contents using an LLM and serves documents, structure, and page co…