Tavily MCP
github.com/tavily-ai/tavily-mcp- Category
- Developer Tools
- Rank
- No. 413Tools index
- Pricing
- Freemium
- Type
- TOOL
- Builder
- tavily-ai
- GitHub
- 2.3k stars
- Added
- Jul 4, 2026
About
Tavily MCP is a production-ready MCP server that gives an AI agent real-time web search, content extraction, site mapping, and crawling through a single Model Context Protocol connection. Built by Tavily, whose search API is purpose-built for LLMs and agents.
What it does
Tavily MCP turns web research into agent-callable tools. An MCP client can search the web, extract page content, map sites, or crawl linked pages. It can run locally through standard input and output, or connect to Tavily’s hosted endpoint.
Why it's ranked here
The tool covers several common research steps behind one agent interface, with useful controls for domains, dates, result depth, crawl scope, and output format. Remote and local connection options broaden its fit. The main reservation is dependence on Tavily’s service, with an API key required for the full toolset.
What's good
Search supports date windows, domain filters, country boosting, exact phrases, images, raw content, and several depth settings. Extraction accepts multiple URLs, markdown or text output, image inclusion, and relevance reranking. Crawl controls expose depth, breadth, and total processing limits. Shared defaults reduce repeated configuration across requests.
Tradeoffs
Full operation depends on Tavily’s hosted APIs rather than a self-contained search or crawling stack. Keyless local mode exposes only search and extraction; mapping and crawling require an API key. The documented remote URL can place that key in a query string, although bearer authentication and OAuth are also available.
How to use it well
Use it when an MCP-capable agent needs current discovery followed by targeted page extraction or bounded site exploration. Set shared defaults for recurring research patterns, and use opaque end-user identifiers if you need per-user analytics. Pair it with separate storage, indexing, graph, or synthesis systems because this project supplies retrieval tools, not those downstream layers.
Technical notes+
src/index.ts implements a TypeScript MCP server using @modelcontextprotocol/sdk, stdio transport, Axios, environment loading, and command-line parsing. It creates one session UUID, sends bearer or keyless access headers, optionally forwards X-Human-Id, parses DEFAULT_PARAMETERS, and registers schemas for search, extraction, crawling, mapping, and research endpoints. package.json defines an ES module npm CLI package and TypeScript build scripts. tsconfig.json targets ES2022 with strict checking and Node16 module resolution. Dockerfile uses a two-stage Node Alpine image and installs production dependencies in the release stage. smithery.yaml declares a stdio launch configuration requiring a Tavily API key.
Observed
- License
- MIT
- Primary language
- TypeScript
- Package surface
- ES module npm CLI package, runnable through NPX
- Interfaces
- Local MCP over stdio and documented remote MCP over HTTP
- Runtime
- Local documentation requires Node.js 20 or higher
- Container support
- Two-stage Node Alpine Docker image
- Authentication
- API key, bearer header, optional OAuth for remote clients, and limited keyless local mode
Read from README.md, package.json, src/index.ts, LICENCE, Dockerfile, smithery.yaml, tsconfig.json.
Tags
Tech Stack
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.