mcp -tavily

This MCP server provides a Tavily-based search tool. It exposes a single tool named 'search' which sends a query to the Tavily API and returns results in text form, including AI responses, URLs, and result titles. When integrated with Claude Desktop or similar clients, you can invoke the tool by asking the client to perform a search, for example: 'Please search for today's events in Kamakura.' The server will call Tavily with your query and depth setting (basic or advanced if supported) and return a structured text response with a list of results and any AI-generated synthesis. The setup is designed for environments where Python-based tooling (uv) runs the tavily-search module, and the client passes your natural language requests to the tool for web-based results.

Key capabilities: search via Tavily API, optional search_depth parameter to control depth, and formatted text responses that include AI commentary, individual result entries, and URLs. The configuration expects an API key to authenticate Tavily requests. Ensure your client can route queries to the tavily-search tool and display the text-based results to the user.

Prerequisites:

• Git installed on your system
• Python environment available (to support the uv runner)
• Tavily API key (for TAVILY_API_KEY)
• Access to clone the MCP server repository

Installation steps:

1. Clone the MCP server repository

git clone https://github.com/Tomatio13/mcp-server-tavily.git
cd mcp-server-tavily

2. Prepare the environment

• Ensure Python is installed (Python 3.x)
• Install the uv runner if not already available in your environment (the setup assumes uv is used to run the server):

# Example: install uv if using pipx/venv (adjust as needed for your setup)
pip install uv        # or appropriate installation method for your environment

3. Configure the Tavily API key

• Obtain your Tavily API key and set it in the environment or config file as shown in the mcp configuration example:

export TAVILY_API_KEY=YOUR_TAVILY_API_KEY

• For Windows PowerShell:

$env:TAVILY_API_KEY = "YOUR_TAVILY_API_KEY"

4. Run the Tavily search MCP server

• Using the uv runner as described in the README configuration:

uv --directory C:\your_path\mcp-server-tavily run tavily-search

• Alternatively, adjust the directory path to where you cloned the repository and ensure the tavily-search entry point is available.

Optional: Docker-based run

• If you prefer Docker, follow the Docker Compose section in the README to run the client and server in containers. This requires Docker and docker-compose installed on your system, and uses the tavily-search server entry within the container setup.

Tips and notes:

• The server relies on Tavily API access; ensure your TAVILY_API_KEY is valid and has needed permissions.
• The mcpServers configuration uses the uv runner with a directory path to your local repository and a command to start tavily-search. Update the path to your actual clone location.
• If you encounter encoding issues, ensure PYTHONIOENCODING is set to utf-8 as shown in the example.
• When integrating with Claude Desktop or similar clients, the client will send natural language queries which are transformed into tool calls (e.g., search) with the required parameters. The output is a plain-text response suitable for display.
• If you see connection or API errors, verify network access to Tavily endpoints and confirm that the API key is correctly configured in the environment.
• For Windows users, ensure the directory paths in the command line are properly escaped and use double backslashes. For example: --directory C:\your_path\mcp-server-tavily