Gene Library Courses Download Pricing Contact Sign in
tavily logo
search Official Website

Tavily MCP Server – Web Search and Extraction

Tavily's official MCP server gives AI agents access to real-time web search, content extraction, crawling, and website mapping. Use it to ground answers in current sources, extract page content, discover site structure, and power research or RAG workflows from MCP-compatible clients.

#web-search#extraction#crawling

Overview

Tavily provides an official MCP server that connects AI agents to Tavily's search, extraction, crawling, and website-mapping APIs. The server is designed for agentic workflows that need current web information, source-grounded research, structured content extraction, or website discovery without building a custom search integration.

What the MCP server enables

Tavily MCP exposes the core Tavily web-intelligence tools to MCP-compatible clients:

  • tavily-search for real-time web search with answer-focused results and source context.
  • tavily-extract for retrieving and extracting useful content from web pages.
  • tavily-crawl for systematically crawling a website.
  • tavily-map for discovering and mapping pages on a website.
  • Optional tool arguments such as query, URLs, depth, topic, time range, included or excluded domains, and result limits where supported by the underlying Tavily APIs.

The server is useful when an agent needs live web grounding, research context, retrieval for RAG workflows, or website-level discovery.

When to use it

Use Tavily MCP when an AI workflow needs fresh information or source-backed web context. Practical examples include researching recent market developments, checking product documentation, extracting data from specific pages, building a site map before crawling, gathering sources for a report, monitoring competitors, or grounding a chatbot's answer with current web content.

It is especially useful for research agents, coding assistants, analyst workflows, RAG pipelines, and applications where stale model knowledge is not enough.

Connection and authentication

Tavily documents a hosted MCP endpoint for remote clients:

https://mcp.tavily.com/mcp/?tavilyApiKey=${TAVILY_API_KEY}

Tavily also distributes an official npm MCP server package:

npx -y @tavily/mcp@latest

Local stdio configuration requires a Tavily API key in the TAVILY_API_KEY environment variable. Tavily's documentation and GitHub repository show setup examples for clients such as Cursor, Claude Desktop, Claude Code, OpenAI Agent Builder, Google ADK, Arcade, Azure, Databricks, and Amazon Bedrock AgentCore integrations.

Key considerations

A Tavily API key is required for authenticated access and should be stored securely, never hard-coded in shared files. Tavily usage is subject to the account's API credits, rate limits, and plan limits. Remote MCP URLs that include an API key should be treated as secrets because the key appears in the connection string. Use domain filters, time ranges, depth limits, and result limits to control cost and scope. For high-stakes research, review returned sources and extraction quality before relying on synthesized conclusions.

Supported Transports

streamable_http

URL: https://mcp.tavily.com/mcp/?tavilyApiKey=${TAVILY_API_KEY}

stdio

Command: npx

Args:

  • -y
  • @tavily/mcp@latest

Frequently Asked Questions

When should an AI agent use the Tavily MCP server?
Use it when a workflow needs current web information, source-grounded research, web page extraction, website crawling, page mapping, competitor monitoring, or retrieval context for an AI or RAG pipeline.
What does the Tavily MCP server add to an AI agent's capabilities?
It gives the agent access to Tavily's live web search, extraction, crawl, and map tools so the agent can gather current sources and structured web context instead of relying only on static model knowledge.
What can an AI agent access or manage through Tavily MCP?
The agent can run Tavily search queries, extract content from URLs, crawl websites, and map website structure. The server is for web-intelligence operations and does not manage unrelated user accounts or local files.
How is authentication configured for the Tavily MCP server?
Tavily requires a Tavily API key. Remote MCP examples pass the key in the hosted URL as tavilyApiKey, while local stdio configuration uses the TAVILY_API_KEY environment variable with the official @tavily/mcp package.
Which transport should be used for the Tavily MCP server?
Use the hosted Streamable HTTP endpoint when the MCP client supports remote servers. Use the official local stdio package with npx -y @tavily/mcp@latest when a client needs a local MCP process.