Give your AI assistant live web search and a crawler

What it does Tavily MCP is a Node-based MCP server that exposes Tavily’s web API as four standard tools—search, extract, map, and crawl—so any MCP-compatible client can read the live web. Your agent gets real-time search results, structured page extraction, site mapping, and systematic crawling instead of relying on static training data. You can run it locally or use Tavily’s hosted remote server and skip installation entirely.

The interesting bit The project leans hard into convenience: the remote server at mcp.tavily.com means zero local dependencies beyond an API key or OAuth flow. The optional TAVILY_HUMAN_ID environment variable, hashed server-side before storage, is a rare privacy-conscious nod to multi-user deployments that need analytics.

Key highlights

• Four MCP tools: tavily-search, tavily-extract, tavily-map, and tavily-crawl
• Fully hosted remote option with OAuth and API-key authentication
• Default request parameters configurable via DEFAULT_PARAMETERS header or env var
• Per-user analytics support via hashed X-Human-Id forwarding
• Ready-made connection guides for Claude Code and Cursor

Verdict Worth a look if you already use Tavily and want your Claude or Cursor instance to stop hallucinating about last year’s documentation. Skip it if you need a self-hosted, offline search backend—this is a commercial-API wrapper, not a search engine in a box.