MCP Server Tavily

What Tomatio13/mcp-server-tavily MCP server does

Tomatio13/mcp-server-tavily MCP server connects an MCP client to the Tavily search API. Its purpose is narrow: it performs web searches from agent requests and presents the returned information as text. A response can contain an AI-generated answer followed by individual search results, including their titles and URLs. The example output also shows a summary field for each result, although that field may report that a summary was not found.

The repository is suitable for agents that need to look up changing information, such as current events or other web content. It does not describe tools for storing data, crawling a site, modifying web pages, or managing search indexes.

How it works

The server exposes one MCP tool named search. The query argument is required and holds the natural-language or keyword search request. search_depth is optional and accepts either basic or advanced, allowing the client to select the documented Tavily search depth.

When the tool is called, the server sends the request to Tavily's search endpoint and formats the response as text. The documented result can include an AI answer plus a numbered list of search results. The example shows an HTTP request to https://api.tavily.com/search and a successful MCP tool response.

Setup and configuration

To run Tomatio13/mcp-server-tavily MCP server, clone the repository and configure an MCP client to launch it with uv. The Claude Desktop example uses uv --directory, followed by the local repository directory, run, and the tavily-search entry point. The repository path in the example must be replaced with the actual checkout location.

Set TAVILY_API_KEY in the server environment. The configuration examples also set PYTHONIOENCODING to utf-8. After saving the Claude Desktop configuration, restart the application. The README gives configuration locations for macOS and Windows.

Cursor can launch the server through a shell script that changes into the repository, exports the environment variables, and runs the same uv entry point. The README also describes a Docker Compose flow: start the compose services, then execute the included client inside the mcp_server container.

Tools and capabilities

• search: searches Tavily using a required query.
• Selects basic or advanced processing through search_depth.
• Returns text containing an AI answer and search result links.
• Supports Claude Desktop and Cursor configurations documented by the repository.

Limitations and notes

Tomatio13/mcp-server-tavily MCP server requires a Tavily API key, so it is not a credential-free search option. The provided material does not specify Tavily quotas, pricing, rate limits, or the repository's software license. Search output depends on Tavily and the current web; the README's example advises checking official websites for up-to-date event information.

The README documents log storage under Claude Desktop's application data directory on Windows. It does not provide equivalent log locations for every client or operating system. The Docker instructions are intended for local environments where Claude Desktop is unavailable and require Docker and Docker Compose.