Tavily MCP Server

by Tavily AI

Tavily MCP Server gives AI agents the ability to search the live web, extract structured content from pages, map entire website architectures, and crawl sites systematically — all through the Model Context Protocol. Rather than relying on static training data, agents connected to Tavily can fetch real-time search results with advanced filtering, pull clean text and metadata from any URL, discover every page on a domain via site mapping, and perform deep multi-page crawls to build comprehensive knowledge bases on the fly. The server exposes four core tools. tavily-search delivers real-time web search with configurable depth, domain filtering, and optional image results. tavily-extract pulls structured data from individual web pages, stripping away boilerplate to return clean, usable content. tavily-map creates a complete structural map of a website, listing all discoverable URLs and their relationships. tavily-crawl systematically explores a site page by page, combining mapping and extraction into a single workflow for thorough data gathering. Deployment is flexible: run locally via npx with a simple API key, connect to the hosted remote server at mcp.tavily.com without installing anything, or use Docker for containerized environments. Authentication supports API keys (via query parameter or Authorization header) and a full OAuth flow for secure, token-based access. Default parameters like search depth, max results, and image inclusion can be set globally through environment variables or HTTP headers, ensuring consistent behavior across all requests. The server is compatible with Claude Desktop, Claude Code, Cursor, and any MCP-capable client, making it one of the most versatile web research tools available for AI agent workflows.

FAQ