Knowledge MCP Server

What is Knowledge MCP Server?

A Model Context Protocol (MCP) server that provides tools for managing and querying knowledge bases through vector databases. It is built on the dynamic-mcp-server framework and is intended for AI models to create, manage, and query knowledge sources using semantic search.

How to use Knowledge MCP Server?

Clone the repository, install dependencies with npm install, configure a .env file with MONGODB_URI, OPENAI_API_KEY, and optionally PORT and HOST, then run with npm run dev (development) or npm start (production). The server exposes tools like add-knowledge, search, use-knowledge-source, and refresh-knowledge-source.

Key features of Knowledge MCP Server

• Vector database integration for storing and querying embeddings
• Document processing pipeline with text chunking and embedding generation
• Dynamic tool registration and management for knowledge sources
• Website crawling and content extraction
• MongoDB integration for metadata and vector storage
• Secure access control with ownership and sharing levels

Use cases of Knowledge MCP Server

• Ingest website content into a searchable knowledge base
• Perform semantic search across document fragments with metadata filtering
• Dynamically create a dedicated tool for querying a specific knowledge source
• Refresh a knowledge source by re‑ingesting its content and updating embeddings

FAQ from Knowledge MCP Server

What runtime and dependencies are required?

Node.js 18 or later, a MongoDB Atlas M10 or higher instance (for vector search), and an OpenAI API key.

How does the server handle access control?

Users own the knowledge sources they create and can share them with others at “read” or “write” access levels.

What knowledge source types are supported?

Currently only “Website” (using a built‑in web crawler); Microsoft OneDrive support is listed as a future implementation.

Which tools does the server provide?

Four tools: add-knowledge (ingest documents), search (semantic query), use-knowledge-source (create a dedicated tool), and refresh-knowledge-source (re‑ingest content).

Where are metadata and vectors stored?

Metadata and vector embeddings are stored in MongoDB via the @llm-tools/embedjs-mongodb library.