CookiesForAI

Tavily MCP server

The Tavily MCP server gives AI agents real-time web search through the Tavily API, plus tools that extract page content, map a site's structure and crawl it.

By Tavily Official · Updated

Connect Tavily

Claude Code, with OAuth sign-in (finish it with /mcp)

Terminal
claude mcp add --transport http tavily https://mcp.tavily.com/mcp

Claude Code, with an API key in the URL

Terminal
claude mcp add --transport http tavily https://mcp.tavily.com/mcp/?tavilyApiKey=<your-api-key>

Cursor, in mcp.json

Config
{"mcpServers": {"tavily-remote-mcp": {"command": "npx -y mcp-remote https://mcp.tavily.com/mcp/?tavilyApiKey=<your-api-key>", "env": {}}}}

Run the server locally (set TAVILY_API_KEY in its environment)

Terminal
npx -y tavily-mcp@latest

Tavily accepts the API key in the URL, as an Authorization: Bearer header, or through OAuth. In Claude Desktop, add the remote URL as a custom connector.

Or paste this into your coding agent: Connect the Tavily MCP server to my coding agent using its documented setup: https://github.com/tavily-ai/tavily-mcp.

What the Tavily MCP server does

The Tavily MCP server connects an AI agent to Tavily, a search API made for language models. Its search tool queries the live web and returns ranked results with content snippets, with options for search depth, number of results, images and time range. Its extract tool pulls the readable content out of one or more URLs, so the agent can read a source after finding it.

Tavily's server also covers whole sites. A map tool returns the structure of a website as a list of its pages, and a crawl tool walks a site from a starting URL and collects content along the way. The source code also registers a research tool for longer multi-step questions. Default parameters such as search depth or result count can be set once for every call, through an environment variable locally or a header on the remote server.

Tavily hosts the server at mcp.tavily.com and supports OAuth sign-in or an API key, so most people connect a URL. The same server runs locally with npx and Node.js 20 or newer. The GitHub repo had 2,415 stars on 2026-10-01.

Tools it gives your agent

  • tavily_search
  • tavily_extract
  • tavily_map
  • tavily_crawl
  • tavily_research

When to use Tavily

  • Your agent needs fresh search results with the page text, not only links.
  • You want to pull the content of a list of URLs into the conversation.
  • You need a list of every page on a documentation site before reading some of them.
  • You want a search provider with a free monthly allowance for a side project.

When to pick something else

  • Use without an account: every call needs a Tavily API key or OAuth sign-in, and the free plan has a monthly credit limit.

What Tavily needs

  • A Tavily account or API key
  • Node.js 20 or newer, only when running the server locally

What the Tavily MCP server costs

Free and open source under MIT; calls spend Tavily API credits, with a free monthly allowance, then pay as you go or a paid plan.

Current prices: www.tavily.com . We do not list prices: vendors change them often, so check the publisher's own page.

Questions people ask

Is the Tavily MCP server free?

The server is free and open source, and Tavily's free Researcher plan includes a monthly allowance of API credits. After that, you pay as you go or move to a monthly plan (pricing page, 2026-10-01).

Do I need an API key for the Tavily MCP server?

Yes, or an OAuth sign-in to a Tavily account. Unlike Exa and Firecrawl, Tavily has no keyless mode.

  • Apify: The Apify MCP server gives AI agents thousands of ready-made scrapers and automation tools from Apify Store, such as Google Maps, Instagram and search result scrapers, which the agent can find and run on demand.
  • Brave Search: The Brave Search MCP server gives AI agents Brave's independent search index: web, news, image, video, local and place search, plus pre-extracted page text for grounding and AI summaries.
  • Exa: The Exa MCP server gives AI agents web search and page reading through Exa's search engine, plus optional advanced filters and Exa Agent runs for multi-step research with structured output.
  • Firecrawl: The Firecrawl MCP server gives AI agents live web access through Firecrawl: search the web, scrape a URL into Markdown or JSON, map and crawl whole sites, and click through pages.