Tavily MCP Server

What is Tavily MCP Server?

A FastAPI-based server that wraps the Tavily API into an MCP (Model Context Protocol) service. It provides endpoints for web search and content extraction, secured by API key authentication. Designed for AI agents and developers who need programmatic access to Tavily's search capabilities via the MCP protocol.

How to use Tavily MCP Server?

Install dependencies with pip install -r requirements.txt, set the API_KEY and TAVILY_API_KEY environment variables, then run uvicorn server:app --host 0.0.0.0 --port 8000 --reload. The server exposes two POST endpoints: /tavily-search and /tavily-extract.

Key features of Tavily MCP Server

• Remote SSE MCP server built with fastapi-mcp
• Tavily web search with configurable depth and topic
• Tavily content extraction from given URLs
• API key authentication for server access
• Customizable search parameters (max results, domains, time range)
• Optional image inclusion and raw content retrieval

Use cases of Tavily MCP Server

• An AI agent performing real‑time web search during a conversation
• Extracting clean text from a list of URLs for content analysis
• Integrating Tavily search into a larger MCP‑based tool ecosystem
• News search with time range and topic filtering
• Domain‑restricted or excluded searches for focused results

FAQ from Tavily MCP Server

What API keys are required?

The server needs two environment variables: API_KEY (for accessing the server's endpoints) and TAVILY_API_KEY (for accessing the Tavily API).

How do I run the server?

After installing dependencies, set the two API keys as environment variables and run uvicorn server:app --host 0.0.0.0 --port 8000 --reload. The server will be available at http://127.0.0.1:8000.

What endpoints does the server provide?

Two POST endpoints: /tavily-search for web search and /tavily-extract for content extraction from URLs. Both require authentication via the server’s API key.

What search parameters can I configure?

Search supports optional parameters: search_depth (basic/advanced), topic (general/news), days, time_range, max_results (default 10), include_images, include_image_descriptions, include_raw_content, include_domains, and exclude_domains.

What extraction parameters are available?

Extraction requires a list of urls. Optional parameters: extract_depth (basic/advanced, default basic) and include_images (default false).