Tavily Search MCP Server

Tavily MCP is an officially maintained Model Context Protocol server from Tavily that gives LLM agents real-time access to the web. It wraps Tavily's search API, which is purpose-built for AI agents, returning clean, ranked, citation-ready content instead of raw HTML. The server is available as a hosted remote endpoint at https://mcp.tavily.com/mcp/ or as a local npm package (tavily-mcp).

The server exposes four tools: tavily-search for live web search with filters like topic, time range, and domain inclusion/exclusion; tavily-extract for pulling clean content from specific URLs; tavily-map for generating a structured map of a website's pages; and tavily-crawl for systematic multi-page exploration. Results can include AI-generated answers, images, and raw content depending on parameters.

Tavily offers a free tier with monthly API credits, making it a common choice for grounding LLM responses in current web data, fact-checking, market research, and competitive intelligence workflows. The server supports both hosted (URL or Bearer token auth) and local (env var) deployment, and works with Claude Desktop, Claude Code, Cursor, and other MCP clients.

Tools

Prerequisites

• A Tavily API key from tavily.com (free tier available)
• Node.js v20+ (only required for local install)
• An MCP-compatible client (Claude Desktop, Claude Code, Cursor, etc.)

Option 1: Remote server (recommended)

Point your MCP client at Tavily's hosted endpoint with your API key:

https://mcp.tavily.com/mcp/?tavilyApiKey=<your-api-key>

For Claude Code:

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

For Claude Desktop, add to claude_desktop_config.json:

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

Option 2: Local install via npx

{
  "mcpServers": {
    "tavily-mcp": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": {
        "TAVILY_API_KEY": "your-api-key-here"
      }
    }
  }
}

Optional env vars:

• DEFAULT_PARAMETERS: JSON string of default tool parameters (e.g. {"include_images": true})
• TAVILY_HUMAN_ID: hashed end-user identifier for per-user analytics

• Ground LLM answers in current web data by running tavily-search before responding to time-sensitive questions
• Fact-check claims by searching for sources and extracting content from specific URLs with tavily-extract
• Build competitive intelligence reports by mapping a competitor's site with tavily-map and crawling key sections
• Aggregate news on a topic over a custom time window using the topic and days parameters of tavily-search
• Scrape documentation or knowledge bases for RAG pipelines using tavily-crawl with path filters

• "Search the web for the latest funding rounds in AI infrastructure this week and summarize the top five."
• "Extract the full article content from these three URLs and compare their stances on the new EU AI Act."
• "Map the site structure of stripe.com/docs and list the top-level sections."
• "Crawl example.com starting from /blog, limit to 50 pages, and pull the titles and publish dates."
• "Find recent reviews of the Sony WH-1000XM6, include images, and give me a synthesized pros and cons list."

• Officially maintained by Tavily with both a hosted remote server and a local npm package
• Search results are LLM-optimized: ranked, deduped, and returned as clean text with optional AI-generated answers
• Four complementary tools (search, extract, map, crawl) cover most agent web-access needs
• Free tier with monthly credits makes it easy to try without payment setup

• Requires a Tavily account and API key; heavy usage moves to paid tiers
• Crawl and map tools consume more API credits than basic search, which can add up quickly
• Results quality depends on Tavily's index and ranking, which is less transparent than running your own search

• Exa MCP Server for neural web search optimized for AI
• Brave Search MCP for independent index web search
• Firecrawl MCP for web scraping and crawling focused workflows