Overview
The MCP Documentation Server is a TypeScript-based MCP server that provides local-first document management and semantic search using embeddings. It exposes a collection of MCP tools and is optimized for performance with on-disk persistence, an in-memory index, and caching. The server also supports AI-powered analysis via Google Gemini AI, enabling advanced document understanding, natural language queries, smart summaries, explanations, and contextual insights (Gemini API key required). Core capabilities include AI-powered search with Gemini for contextual insights, traditional semantic search using chunk-based embeddings plus an in-memory keyword index, and context window retrieval to assemble surrounding chunks for richer LLM responses. For performance, it offers O(1) document lookup through a DocumentIndex, an LRU EmbeddingCache to avoid recomputing embeddings, parallel chunking and batch processing, and a streaming file reader for large files. Data storage is local-only, with all data residing under ~/.mcp-documentation-server/ (data/ for JSON documents and uploads/ for imports). Exposed MCP tools cover document management, file processing, and search; configuration is environment-variable driven (e.g., MCP_EMBEDDING_MODEL, GEMINI_API_KEY).
Features
AI-Powered Document Analysis (Gemini AI)
Enhanced with Google Gemini AI for advanced document analysis, contextual understanding, and insights (requires GEMINI_API_KEY).
Traditional Semantic Search
Chunk-based search using embeddings plus an in-memory keyword index for fast, accurate retrieval.
Context Window Retrieval
Fetch surrounding chunks to provide richer context for LLM responses.
O(1) Document Lookup
DocumentIndex enables instant document lookup for quick access.
LRU Embedding Cache
EmbeddingCache avoids recomputing embeddings to speed up repeated queries.
Parallel Chunking & Batch Processing
Parallel processing accelerates ingestion of large documents.
Streaming File Reader
Reads large files with low memory usage via streaming.
Local-Only Storage with Disk Persistence
All data remains locally in ~/.mcp-documentation-server/ with on-disk persistence.
Who Is This For?
- Developers:Build MCP-driven apps and workflows with local-first document management.
- Knowledge workers:Search, summarize, and relate content across documents using AI-enabled queries.
- Content teams:Organize and retrieve information from local files efficiently.
