Introduction

ContextMCP is a self-hosted alternative to Context7. It allows your AI Agents to search across your documentation, repositories and API references using the Model Context Protocol or REST API endpoint.

Features

• Self-Hosted - Your docs stay on your infrastructure
• Semantic Search - Natural language queries with vector embeddings
• MCP Native - Works with Cursor, Windsurf, Claude Desktop out of the box
• Fast - Cloudflare Workers edge deployment
• Fully Configurable - Simple YAML configuration, no code changes needed
• Easily Extendable - Add new parsers and chunkers or modify existing ones with ease

How It Works

ContextMCP follows a simple workflow:

1. Configure - Define your documentation sources in config.yaml
2. Index - Run npm run reindex to fetch, parse, and embed your docs
3. Deploy - Deploy the Cloudflare Worker to serve search requests
4. Connect - Configure your AI Agents to use the MCP server

The system:

• Fetches documentation from GitHub, GitLab, local files, or URLs
• Parses MDX, Markdown, OpenAPI, or HTML formats
• Chunks content intelligently for optimal retrieval
• Generates embeddings with OpenAI, Gemini, Cohere, Voyage, or local Ollama
• Stores vectors in Pinecone for fast semantic search
• Serves results via MCP protocol and REST API

Architecture

Next Steps

Ready to get started? Follow the Quick Start guide to set up ContextMCP in minutes.

Want an end-user chat UI on top of your docs? Pair ContextMCP with ContextChat — a drop-in, embeddable "Ask AI" widget that streams cited answers from your /search endpoint.