andrea9293/mcp-documentation-server

What it solves

It provides a local-first document management and semantic search system specifically designed for AI coding agents. It eliminates the need for external databases or cloud APIs, preventing vendor lock-in and ensuring data privacy by keeping knowledge bases offline.

How it works

The server uses the Model Context Protocol (MCP) to expose document management and search tools to AI agents. It employs a "parent-child chunking" strategy where documents are split into large context-preserving parent chunks and smaller child chunks for precise vector search. Search is powered by Orama (a local vector DB) and Transformers.js for local embeddings. It also includes a built-in web dashboard for manual document management and an optional integration with Google Gemini for advanced AI-powered analysis.

Who it’s for

Developers using AI coding agents (such as Claude Code, Cursor, or Gemini CLI) who want to give their agents access to a private, local knowledge base of documentation, PDFs, and markdown files.

Highlights

• Fully Offline: Uses local embeddings and a local vector database for privacy and speed.
• Web Dashboard: Includes a full browser-based UI to upload, browse, and manage documents.
• Hybrid Search: Combines full-text and vector similarity search.
• REST API Access: Exposes tools via REST to avoid bloating agent conversation contexts with large MCP schemas.
• Flexible File Support: Supports .txt, .md, and .pdf uploads via drag-and-drop.