ContextMCP

What is ContextMCP?

ContextMCP is a self-hosted MCP server that indexes documentation from multiple sources and serves it via the Model Context Protocol (MCP) and a REST API. It creates a searchable knowledge base that AI assistants can query, making documentation accessible to LLMs.

How to use ContextMCP?

Scaffold a new project with npx contextmcp init, then configure your Pinecone API key and an embedding provider key (OpenAI or Gemini) in the .env file. Edit config.yaml to specify your documentation sources, run npm run reindex to index, and deploy the Cloudflare Worker with npm run deploy. The server is then ready to accept MCP queries.

Key features of ContextMCP

• Self‑hosted MCP server for documentation indexing
• Supports MDX/JSX, plain Markdown, and OpenAPI/Swagger specs
• Chunks, embeds, and stores content in Pinecone vector database
• Searchable via both MCP and REST API
• Works with OpenAI or Gemini for embeddings
• Includes a CLI scaffolding tool (npx contextmcp)

Use cases of ContextMCP

• Allow AI assistants to query your product documentation directly
• Add a searchable knowledge base to documentation sites like Mintlify or Docusaurus
• Power a drop‑in “Ask AI” chat widget using the companion ContextChat
• Provide API reference search via OpenAPI specs

FAQ from ContextMCP

What documentation sources are supported?

ContextMCP can parse MDX/JSX (e.g., Mintlify, Fumadocs, Docusaurus), plain Markdown files (READMEs, CHANGELOGs), and OpenAPI/Swagger specifications.

Which embedding providers are available?

You can use either OpenAI (by setting OPENAI_API_KEY) or Gemini (by setting GEMINI_API_KEY) as the embedding provider in the .env file.

Where is the indexed data stored?

All embeddings and content chunks are stored in a Pinecone vector database. You need to provide your own PINECONE_API_KEY.

What are the system requirements?

Node.js 18 or later is required to run the CLI and build the project.

Can I add a chat UI to my docs?

Yes. Pair ContextMCP with ContextChat, a single Cloudflare Worker that queries the /search endpoint and streams answers with inline citations. It can be added to any site with one <script> tag.