mcp-tavily-search

The mcp-tavily-search MCP server exposes Tavily’s search capabilities to your LLM workflows. It supports three MCP tools: tavily_search for performing advanced Tavily web searches with optional domain filtering, context generation via tavily_get_search_context for RAG-style use, and tavily_qna_search for direct question answering. Each tool accepts a configurable set of parameters, including search depth, topic, time ranges, and caching options, and can return results in several formats (text, json, or markdown) with optional AI-generated summaries. To use the server, run it through your MCP client (for example via npx as shown in the configuration) and pass the appropriate tool name and arguments to your agent or orchestrator. Ensure you provide your Tavily API key via the TAVILY_API_KEY environment variable when launching the server.

Prerequisites:

• Node.js (recommended latest LTS) and npm/yarn installed on your system
• Access to the MCP environment where you will run the server
• Tavily API key for Tavily Search API

Installation steps:

1. Ensure Node.js and npm are installed

   • macOS/Linux: https://nodejs.org/
   • Windows: install via the Node.js installer

2. Install or run via npx (no local install required):

# Using npx as described in the MCP configuration
npx -y mcp-tavily-search

3. If you prefer a local setup, install the package (if published) and run it directly:

npm install -g mcp-tavily-search
mcp-tavily-search

4. Provide your Tavily API key in the environment:

export TAVILY_API_KEY=your-tavily-api-key
# or on Windows
set TAVILY_API_KEY=your-tavily-api-key

5. Integrate with your MCP client by configuring an mcpServers entry (see mcp_config) and launching the server as part of your MCP workflow.

Notes and tips:

• This repository is no longer actively maintained; functionality is superseded by mcp-omnisearch for a unified toolset.
• Ensure your Tavily API key is kept secure and not checked into source control.
• If you’re using domain filtering, you can specify include_domains or exclude_domains to tailor sources.
• The server supports multiple response formats and can return cached results with TTL control; tune cache_ttl as needed for your use case.
• If you encounter issues, verify that the Tavily API endpoint is reachable and that the API key has the required permissions.
• When deploying in WSL or cross-platform environments, ensure the shell initialization (e.g., nvm) is available if using npx within scripts.