Search and extract are the pair most people want, and having map and crawl next to them means "read this whole site" is one server rather than three. The reason to pick this implementation is deployment: it runs as a shared SSE service in a container, so several clients use one process and one API key instead of each spawning their own.
A FastMCP implementation of the Tavily tools that runs either as a local stdio process or as a standalone SSE service, including a Docker image. Same tools either way; the difference is whether one client launches it or several connect to it.
- `tavily-search` for real-time web search
- `tavily-extract` for pulling content out of a specific page
- A mapping tool that builds a structured map of a website
- A crawler that walks a site systematically
- Two entry points from one install: `tavily-mcp-sse` and `tavily-mcp-stdio`
- A Dockerfile, so the SSE server can run as a container on port 8000
A Tavily account and API key in `TAVILY_API_KEY`, plus git and `uv`; Docker only if you want the containerised SSE server. Copy `.env.example` to `.env` and put the key there, then `uv run --env-file .env tavily-mcp-sse`. `TAVILY_MCP_PORT` changes the SSE port. Version 0.1.0.