Tavily MCP 🔌 MCP server Freemium

What it does

Tavily MCP connects AI agents to Tavily's search API, which is built for LLMs. It gives the agent real-time web search, content extraction from pages, site mapping and crawling. A coding agent can use it to find a new release, a known bug or an API's latest parameters, then pull the page content into context. You can connect to Tavily's hosted remote server or run the open-source server locally through npx.

Tools it gives your agent

• tavily_search: real-time web search with options for depth, result count and images
• tavily_extract: pulls clean content from one or more URLs
• tavily_map: builds a structured map of a website's pages
• tavily_crawl: explores a site systematically and gathers content
• tavily_research: a research tool for multi-step questions
• A feedback tool for reporting result quality

Works with

The remote server at https://mcp.tavily.com/mcp works with any client that supports remote MCP. The README has setup steps for Claude Code and Cursor. The local server needs Node.js 20 or newer and runs over stdio, so Claude Desktop, Codex, Gemini CLI, VS Code and other clients can start it too.

How to connect

Claude Code with OAuth, then run /mcp to sign in:

claude mcp add --transport http tavily https://mcp.tavily.com/mcp

Local:

{ "mcpServers": { "tavily-mcp": { "command": "npx", "args": ["-y", "tavily-mcp@latest"], "env": { "TAVILY_API_KEY": "your-api-key-here" } } } }

Cost and limits

The server is free, but every call uses credits from your Tavily account. At the time of writing (September 2026), the Free plan includes 1,000 API credits a month with no card required. Pay-as-you-go costs $0.008 per credit, and paid plans add more monthly credits. Advanced searches, extraction and crawls use more credits than a basic search, so check Tavily's pricing page for the exact rates.

Safety notes

The README allows passing your API key in the URL (?tavilyApiKey=), but that tends to leak into shell history and shared config files. Use OAuth or an Authorization: Bearer header instead. After OAuth sign-in, a key named mcp_auth_default in your account decides which key is billed. Crawls can use up credits quickly, so set limits on crawl depth and result counts.