Tavily MCP Server
Connect your AI agents to Tavily's web search, content extraction, and crawling via the official remote MCP server.
Updated
What is Tavily MCP?
The Tavily MCP server is a Model Context Protocol server that connects AI agents and assistants to Tavily's AI-optimized search API — giving them real-time web search, content extraction, crawling, and site mapping as callable tools.
Tavily runs the server as an officially hosted remote endpoint at https://mcp.tavily.com/mcp/, so there's nothing to install or self-host for the typical setup. You authenticate either with a Tavily API key (passed in the URL query string or an Authorization bearer header) or via a standard OAuth flow, which lets the client connect without exposing your key in config files. An open-source local server (npx -y tavily-mcp@latest) is also published for self-hosted setups.
Because search results are returned as clean, LLM-ready content rather than raw HTML, Tavily is a common building block for research agents, RAG pipelines, and any workflow that needs fresh information from the open web.
Tools the Tavily MCP server exposes
| Tool | What it does |
|---|---|
| tavily-search | Real-time web search returning AI-ready, LLM-optimized results. |
| tavily-extract | Extract clean, structured content from one or more web page URLs. |
| tavily-crawl | Systematically crawl a website to gather content across its pages. |
| tavily-map | Build a structured map of a website's URL/page structure. |
Connect the Tavily MCP server
Claude Code
- 1
Add the remote server
Run: claude mcp add tavily --transport http https://mcp.tavily.com/mcp/
- 2
Authenticate
On first use, complete the OAuth flow in your browser (or append ?tavilyApiKey=<your-api-key> to the URL to use an API key instead).
- 3
Verify
Run /mcp in Claude Code to confirm the tavily server is connected and its tools (tavily-search, tavily-extract, tavily-crawl, tavily-map) are listed.
Cursor
- 1
Open MCP settings
Edit ~/.cursor/mcp.json (or the project .cursor/mcp.json).
- 2
Add the server
Add an entry under mcpServers using the mcp-remote bridge: "tavily": { "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.tavily.com/mcp/?tavilyApiKey=<your-api-key>"] }
- 3
Reload and verify
Reload Cursor, then check Settings > MCP to confirm Tavily's tools are available to the agent.
Sample use cases
An agent searches the web for the latest news about a competitor and compiles a daily briefing
The team stays informed on competitive moves without manual research.
A research agent queries Tavily for primary sources to fact-check claims in a draft blog post
Published content is backed by verified, up-to-date references.
An agent monitors industry keywords weekly and flags new high-authority articles for link-building outreach
Faster discovery of backlink opportunities from fresh, relevant content.
A scheduled agent searches for mentions of the company brand across the web and alerts the PR team to new coverage
Real-time brand monitoring without paying for a dedicated media tracking tool.
An agent enriches CRM leads by searching for each company's recent press releases and funding rounds
Sales reps get richer context before outreach calls.
A content agent uses Tavily to find statistical data and expert quotes for inclusion in long-form articles
Higher-quality content with authoritative data points, produced faster.
An agent searches for product reviews and user feedback across forums and aggregates sentiment insights
Product teams get a consolidated view of public sentiment without manual scraping.
Security & permissions
Tavily's remote MCP server supports OAuth authentication, so the client connects through a standard browser-based authorization flow without storing your API key in config files. With OAuth, Tavily selects an API key named mcp_auth_default from your dashboard (personal keys take priority over team keys). Alternatively you can authenticate with a raw API key via the URL query string or an Authorization: Bearer header.
The granted access is scoped to Tavily's own search/extract/crawl/map capabilities — the agent can issue web requests and consume your account's API credits, but it does not gain access to other accounts or systems. Treat your Tavily API key as a secret and prefer OAuth where the client supports it.
Using Tavily MCP with Gamut
In Gamut, you add Tavily as a tool on an agent and Gamut manages the OAuth connection to the hosted remote server, so the agent can call tavily-search, tavily-extract, tavily-crawl, and tavily-map as part of an automated, event- or schedule-triggered workflow. For example, a scheduled research agent can search the web for fresh developments on a topic, extract the most relevant pages into clean text, and pass that into a downstream summary or report step — all without a human in the loop. Usage draws on your Tavily API credits.
Frequently asked questions
What kind of searches can the Tavily MCP server perform?
Tavily provides AI-optimized web search that returns clean, structured results. Agents can run keyword searches, filter by recency or domain, and receive extracted content snippets rather than raw HTML.
How does Tavily differ from a standard web search API?
Tavily is purpose-built for AI agents. It returns pre-extracted, relevant content passages instead of a list of links, reducing the need for agents to fetch and parse individual web pages.
Can Tavily search results be filtered by date or source?
Yes. The API supports time-range filters and domain inclusion or exclusion lists, so agents can target recent content from specific sources.
Is Tavily suitable for real-time monitoring workflows?
Yes. Agents can run scheduled Tavily searches at regular intervals to detect new mentions, articles, or data, making it effective for brand monitoring and competitive intelligence.