--- title: "Tavily MCP Server: Search, Extract, Map and Crawl the Web" url: "https://mcpshelf.co/servers/tavily" description: "Tavily's official server gives an assistant web search, page extraction, site maps and crawling through the Tavily API. Hosted URL, npx setup and defaults." updated: "2026-10-27" site: "MCP Shelf" --- # Tavily MCP > Tavily's search, page extraction, site mapping and crawling for agents. ## At a glance Tavily is a search API built for AI agents, and this is the company's own server. Its search tool returns cleaned-up results with optional images and answer text, the extract tool pulls the readable content out of given pages, map lists the pages on a site, crawl walks a site and returns what it finds, and research gathers several sources into one written answer. You need a Tavily API key, which you can get by signing up for a free account. Tavily also hosts the server at mcp.tavily.com, which you can sign into with OAuth instead of pasting a key. | | | | --- | --- | | Made by | Tavily | | GitHub | https://github.com/tavily-ai/tavily-mcp | | Licence | MIT | | npm package | [tavily-mcp](https://www.npmjs.com/package/tavily-mcp) 0.2.22 (5 Aug 2026) | | GitHub stars | 2.5K | | Last commit | 22 Oct 2026 | | Tools | 5 | | Runs on | Your computer or hosted | | Install check | Started in our test | | Category | [Browser and web data](https://mcpshelf.co/categories/browser-and-web) | **Our check:** We ran this exact config on 27 Oct 2026: it started and answered with 5 tools. ## Tools We listed 5 tools on 27 Oct 2026. 0 only read; the rest can change things or run commands. | Tool | Kind | What it is for | | --- | --- | --- | | `tavily_search` | changes | Searches the web and returns snippets and links, with options for depth, topic, dates and images. | | `tavily_extract` | changes | Pulls the readable content out of one or more URLs as Markdown or text. | | `tavily_crawl` | changes | Crawls a site from a URL and returns the content of the pages it reaches. | | `tavily_map` | changes | Lists the pages on a site, starting from a URL. | | `tavily_research` | changes | Researches a question across many sources and writes up the answer. | ### How clearly it describes its tools 5 of 5 tools have a description and 0 of 5 say whether they change things. The tool list takes at least ~400 tokens of context. No tool says whether it only reads or can change things, so your client treats them all alike: keep approvals on. ## Set it up ### Claude Desktop Where: claude_desktop_config.json (Settings → Developer → Edit Config) ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp@latest" ], "env": { "TAVILY_API_KEY": "YOUR_API_KEY" } } } } ``` ### Claude Code Where: Your terminal, in the project folder ```bash claude mcp add-json tavily '{"command":"npx","args":["-y","tavily-mcp@latest"],"env":{"TAVILY_API_KEY":"YOUR_API_KEY"}}' ``` ### Cursor Where: ~/.cursor/mcp.json, or .cursor/mcp.json in a project ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp@latest" ], "env": { "TAVILY_API_KEY": "YOUR_API_KEY" } } } } ``` ### VS Code Where: .vscode/mcp.json in your project ```json { "servers": { "tavily": { "type": "stdio", "command": "npx", "args": [ "-y", "tavily-mcp@latest" ], "env": { "TAVILY_API_KEY": "YOUR_API_KEY" } } } } ``` ### Windsurf Where: ~/.codeium/windsurf/mcp_config.json ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp@latest" ], "env": { "TAVILY_API_KEY": "YOUR_API_KEY" } } } } ``` ### ChatGPT Where: Not supported ChatGPT only connects to remote servers by URL. This one runs on your computer, so use Claude, Cursor or VS Code. Set DEFAULT_PARAMETERS in env to a JSON object, such as {"search_depth": "advanced", "max_results": 10}, to change what every search asks for. ### Remote option Hosted by Tavily. Sign in with OAuth, or add ?tavilyApiKey= with your key to the address. ``` https://mcp.tavily.com/mcp/ ``` ## Settings | Setting | Kind | Required | What it does | | --- | --- | --- | --- | | `TAVILY_API_KEY` | env | yes | Your key from app.tavily.com. Example: `tvly-...` | | `DEFAULT_PARAMETERS` | env | no | A JSON object of defaults applied to every search. The hosted server takes the same as a header. Example: `{"search_depth": "advanced", "max_results": 10}` | | `TAVILY_HUMAN_ID` | env | no | An ID for the person the requests are for, sent to Tavily for per-user analytics. Tavily hashes it before storing it. | | `?tavilyApiKey=` | config | no | Your key as a query string on the hosted URL, for clients without OAuth or header support. Example: `https://mcp.tavily.com/mcp/?tavilyApiKey=tvly-...` | ## What it can touch **It can read** - The public web, through Tavily's search, extraction and crawling **It can change** - Your Tavily credit balance **Safer setup:** It does not change anything outside your Tavily account. Keep crawl limits low, since large crawls use more credits. ## Before you start - A Tavily API key - Node.js 20 or newer, for the local version ## Good for - Giving a chat or coding assistant up-to-date web search - Reading the text of a list of pages without a browser - Mapping or crawling a documentation site ## If it does not work **Calls fail with an authentication error** Check TAVILY_API_KEY in the env block, or the key in the hosted URL. **The OAuth sign-in uses the wrong key** Name the key you want mcp_auth_default in your Tavily account. A personal key with that name wins over a team key. **OAuth gets stuck after a change** Clear the stored tokens with rm -rf ~/.mcp-auth and sign in again. ## Similar servers | Server | Tools | Runs on | Stars | | --- | --- | --- | --- | | [DuckDuckGo Search MCP](https://mcpshelf.co/servers/duckduckgo) | 3 | Your computer | 1.6K | | [Brave Search MCP](https://mcpshelf.co/servers/brave-search) | 8 | Your computer | 1.5K | | [Playwright MCP](https://mcpshelf.co/servers/playwright) | 25 | Your computer | 38.7K | | [Firecrawl](https://mcpshelf.co/servers/firecrawl) | 28 | Your computer | 7.7K | ## Questions ### Is it free? The server is MIT licensed. Tavily's API has a free tier, with paid plans for more use. ### Local or hosted? The tools are the same. The hosted server at mcp.tavily.com needs no Node.js and supports OAuth sign-in. ### Tavily or Exa or Firecrawl? All three search and read the web for agents. Tavily keeps it to five general tools; Firecrawl goes deeper on scraping and Exa on semantic search. ### Is this the official server? Yes. Tavily publishes it in tavily-ai/tavily-mcp. Full docs on GitHub: https://github.com/tavily-ai/tavily-mcp --- From [MCP Shelf](https://mcpshelf.co/): A short shelf of MCP servers that are worth installing. Index for agents: https://mcpshelf.co/llms.txt