Tavily MCP Server

Tavily MCP server provides production-ready AI-optimized web search with real-time extraction, site mapping, and intelligent crawling for LLM agents.

June 10, 2026
MCP ServerSearch & RetrievalTavily MCP Server
GitHub starsnpm versionnpm downloads

Overview

The Tavily MCP Server is an AI-optimized search engine built specifically for LLM agents. Unlike traditional search APIs, Tavily returns structured, cleaned, and summarised content designed to be consumed by language models. It powers agentic search in LangChain, CrewAI, AutoGen, and OpenAI Agents SDK.

Official Server:

Developed and maintained by Tavily. Remote MCP available at https://mcp.tavily.com.

Key Features

🔍

AI-Optimised Search

Returns structured, cleaned content formatted specifically for LLM consumption

📄

Intelligent Extraction

Extract structured data from web pages with AI-powered content parsing

🗺️

Site Mapping

Create structured maps of websites to understand their information architecture

🕷️

Web Crawling

Systematically explore websites with configurable depth and scope

🔄

Real-Time Search

Live web search with fresh results, not cached or stale data

🔑

OAuth Authentication

Secure remote MCP access with OAuth flow, no API key in config files

Available Tools

Quick Reference

ToolPurposeCategory
tavily_searchReal-time web search with configurable depthSearch
tavily_extractExtract structured content from URLsExtraction
tavily_mapBuild a structured map of a websiteMapping
tavily_crawlSystematic multi-page web crawlingCrawling
tavily_researchDeep research with multi-source synthesisResearch

Detailed Usage

tavily_search

Perform a real-time web search with AI-optimised results. Supports basic and advanced search depth.

{
  "query": "latest developments in AI agent frameworks 2026",
  "search_depth": "advanced",
  "max_results": 10,
  "include_images": true
}
tavily_extract

Extract clean, structured content from one or more URLs. Returns markdown-ready text.

{
  "urls": ["https://docs.anthropic.com/en/docs/claude-code"],
  "extract_depth": "advanced",
  "include_images": false
}
tavily_map

Create a structured sitemap of a website showing page hierarchy and relationships.

{
  "url": "https://docs.modelcontextprotocol.io",
  "max_depth": 3
}
tavily_crawl

Systematically crawl multiple pages starting from a root URL with depth control.

{
  "url": "https://docs.modelcontextprotocol.io",
  "max_depth": 2,
  "limit": 50
}
tavily_research

Perform deep research on any topic by gathering and synthesising information from multiple sources. Rate limit: 20 requests per minute.

{
  "input": "Compare the pricing and features of Tavily, Exa, and Brave Search APIs for AI agent integration",
  "model": "auto"
}

Installation

{
  "mcpServers": {
    "tavily": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.tavily.com/mcp/?tavilyApiKey=your-api-key"
      ]
    }
  }
}

Default Parameters

Set default search behaviour with the DEFAULT_PARAMETERS environment variable:

{
  "mcpServers": {
    "tavily": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": {
        "TAVILY_API_KEY": "your-api-key",
        "DEFAULT_PARAMETERS": "{\"include_images\": true, \"search_depth\": \"advanced\", \"max_results\": 15}"
      }
    }
  }
}

API Key Required:

Get a free Tavily API key from app.tavily.com. Free tier includes 1,000 searches/month.

Security

  • API key is the sole authentication mechanism — never commit it to source control
  • Remote MCP supports OAuth for enhanced security (no key in config files)
  • All traffic encrypted via HTTPS
  • human_id parameter hashed server-side (SHA-256) before storage

Sources