The trade against a hosted documentation service is plain: more setup - a vector store, an embedding key, a Worker deploy - in exchange for owning the pipeline and the data. The reindex is a command you run, which means freshness is your responsibility rather than something that happens invisibly.
A scaffolding tool rather than a running service. It generates a project that parses your documentation, chunks and embeds it, stores it in Pinecone, and deploys a Cloudflare Worker that answers MCP and REST queries against it.
- A project scaffolded by one command, with a config file listing the documentation sources to index
- Three parsers: MDX and JSX documentation as produced by Mintlify, Fumadocs or Docusaurus; plain Markdown for READMEs and changelogs; and OpenAPI or Swagger specs for API reference
- Chunking sized for search rather than for reading, and embeddings from OpenAI or Gemini
- A search endpoint served from your own Cloudflare Worker, reachable over MCP or as REST
- A companion chat widget that queries the same endpoint and streams answers with inline citations
Node.js 18+. Scaffold with npx contextmcp init, then supply two keys in the .env: a Pinecone API key for the vector store, and an embedding provider key - OpenAI or Gemini depending on the provider you choose. List your documentation sources in config.yaml, run the reindex, then deploy the Cloudflare Worker. The licence is Apache 2.0.
One command plus a key — npx contextmcp init my-docs-mcp, then supply credentials