The Tavily MCP server, maintained by Tavily, is a search engine built for AI agents. Beyond real-time web search it can extract clean content from specific URLs, crawl a site following links, and map a site's structure — the toolkit a research or fact-checking agent needs to gather and read sources on its own.
Bring your own
Claude Sonnet 4.5
MCP Playground runs 40+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.
A Tavily API key from app.tavily.com — the free tier gives 1,000 searches a month. Deploy a sandboxed hosted instance from the MCP Playground catalog and paste the key at deploy time.
How models use it and what it is built for.
The Tavily MCP server wraps the Tavily API as four MCP tools. Models can run a real-time web search tuned for LLM consumption, extract the main content from one or more URLs as clean markdown, crawl a site by following links to a set depth, and map a site's URL structure without fetching every page. It runs as a Node.js stdio process configured with a Tavily API key, whose free tier includes 1,000 searches a month. Because it combines search with extraction and crawling, a single Tavily connection can power an end-to-end research loop: find sources, open them, and pull structured content back.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Search for the three most-cited papers on MCP security and extract the abstract from each.
Crawl the docs at modelcontextprotocol.io two levels deep and list every page about transports.
Extract the full article at this URL as clean markdown so I can quote it.
Map the structure of example.com and tell me which sections have a pricing page.
This is not a single-model product: you get the same MCP connection with 40+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Tavily
Claude Sonnet 4.5
Claude Sonnet 4.5 orchestrates the search → extract → crawl loop well and knows when each tool applies. Use Haiku 4.5 for single search-and-extract calls.
Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 40+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
It is the official MCP server from Tavily, a search engine designed for AI agents. It exposes search, extract, crawl and map tools so a model can find sources, read them and traverse sites through the Model Context Protocol.
Four tools: tavily-search for real-time web search, tavily-extract to pull clean markdown from specific URLs, tavily-crawl to follow links across a site to a set depth, and tavily-map to map a site's URL structure.
Brave returns ranked links; Perplexity returns a finished cited answer. Tavily sits in between and adds extraction and crawling, so it is the best fit when an agent needs to open specific pages and pull structured content, not just get a list or a summary.
Yes. Tavily's free plan includes 1,000 searches per month, which is enough to evaluate the server and run light research workloads. Higher volumes need a paid plan, but the MCP tools are identical.
Open MCP Agent Studio, deploy the hosted Tavily server from the catalog, paste your Tavily API key, pick a model and ask it to research something. You can watch each tavily-search and tavily-extract call and its output live.
Brave Search
Give an AI agent live web and local search via the Brave Search API.
Perplexity
Get cited, web-grounded answers from an AI agent via Perplexity Sonar.
Wikipedia
Ground AI answers in verifiable Wikipedia articles — no API key needed.
Hacker News
Search and read Hacker News stories, comments and users from an AI agent.
