tavily-search

This MCP server integrates Tavily Search to provide web search capabilities tailored for LLM workflows. It exposes a tavily_search tool that can perform web searches with configurable depth, topic, and time range, and can optionally include images, image descriptions, short LLM-generated answers, raw HTML, and domain filters. You can run the server via Node.js (dist/index.js) or via SSE transport (dist/sse.js) and connect it to Claude Desktop or other MCP clients. To use, supply a query and optional parameters such as search_depth, topic, days, time_range, max_results, and various include_* flags to customize the returned content. The server expects your Tavily API key via the TAVILY_API_KEY environment variable.

Example capabilities:

• tavily_search: performs a web search optimized for LLMs
• Optional outputs: include_images, include_image_descriptions, include_answer, include_raw_content
• Domain filtering: include_domains and exclude_domains to refine results

Prerequisites:

• Node.js and npm installed on your machine
• Access to Tavily API key (TAVILY_API_KEY)
• Basic familiarity with running MCP servers and editing Claude/Desktop config

Install steps:

1. Clone the repository:

git clone https://github.com/apappascs/tavily-search-mcp-server.git

2. Install dependencies and build:

cd tavily-search-mcp-server
npm install
npm run build

3. Set up environment variable with your Tavily API key:

export TAVILY_API_KEY=your_api_key_here

4. Run the server (example with Node):

node dist/index.js

5. If you prefer SSE transport, run:

node dist/sse.js

6. (Optional) Docker: build and run the image following the Docker section in the README.

Prerequisites for Claude/Desktop integration are covered in the README; you’ll typically point the MCP config to the node application path or run via Docker as shown.

Environment variables and secrets:

• Always provide API keys via environment variables (not hard-coded in commands).
• For Docker usage, prefer passing secrets with -e TAVILY_API_KEY=$TAVILY_API_KEY to avoid exposing keys. Transport options:
• stdio transport uses dist/index.js
• SSE transport uses dist/sse.js and can be configured with a port (e.g., 3001) in Claude/Desktop config If you modify path locations, ensure the args point to the correct built distribution (dist/index.js or dist/sse.js). Common issues:
• Missing TAVILY_API_KEY leads to authentication errors with Tavily API
• Incorrect path to dist files results in module not found errors
• When using Docker, ensure you’ve built the image and that volume mappings reflect your local repo path.