Tavily Search MCP Server

Overview

Tavily Search is a Model Context Protocol (MCP) server that provides access to Tavily's powerful search API. This integration allows AI assistants to perform web searches and extract content from URLs with various configuration options.

Prerequisites

Before using Tavily Search, you'll need:

1. Node.js 16 or higher
2. A Tavily API key (obtain one from Tavily's website)

Installation

Using Smithery (Recommended)

The easiest way to install Tavily Search is via Smithery:

npx -y @smithery/cli install @kshern/mcp-tavily --client claude

Manual Installation

You can install the package directly:

npm install @mcptools/mcp-tavily

Or use it with npx without installation:

npx @mcptools/mcp-tavily

Configuration

To use Tavily Search with your AI assistant, add the following configuration to your MCP setup:

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

Replace your-api-key with your actual Tavily API key. Alternatively, you can set the TAVILY_API_KEY environment variable before running the server.

Search Options

All search tools support the following options:

• searchDepth: "basic" or "advanced" - Controls the depth of the search
• topic: "general", "news", or "finance" - Specifies the search topic category
• days: Number of days to search back
• maxResults: Maximum number of results to return
• includeImages: Boolean to include images in results
• includeImageDescriptions: Boolean to include image descriptions
• includeAnswer: Boolean to include an answer in results
• includeRawContent: Boolean to include raw content
• includeDomains: Array of domains to include in search
• excludeDomains: Array of domains to exclude from search
• maxTokens: Maximum number of tokens in the response
• timeRange: "year", "month", "week", "day" (or "y", "m", "w", "d") - Time range for search

Extract Options

The extract tool supports these options:

• extractDepth: "basic" or "advanced" - Controls the depth of extraction
• includeImages: Boolean to include images in extracted content

Response Format

All tools return responses in a standardized format:

{
  "content": [
    {
      "type": "text",
      "text": "Result content here"
    }
  ]
}

For search results, each item typically includes:

• Title
• Content summary
• URL

For extracted content, each item includes:

• URL
• Raw content
• Failed URLs list (if any)

Debugging

For development and debugging, you can use the MCP Inspector, which provides a user interface for testing tool calls, viewing server responses, and monitoring server state.