# Tavily > AI-optimized web search with cleaned, citation-ready results. [Canonical HTML page](https://top-mcps.com/mcp/tavily) · [server.json](https://top-mcps.com/mcp/tavily.json) · [methodology](https://top-mcps.com/about/methodology) ## Install ### Claude Desktop — `claude_desktop_config.json` Paste under mcpServers. Fully quit and reopen Claude after editing. ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### Claude Code — `CLI or .mcp.json` Run from your repo. Commit .mcp.json to share with your team. ```shell # export TAVILY_API_KEY=tvly-your_api_key claude mcp add tavily -- npx -y tavily-mcp ``` ### Cursor — `.cursor/mcp.json` Global path: ~/.cursor/mcp.json. Reload window after editing. ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### VS Code — `.vscode/mcp.json` VS Code uses the "servers" key (not "mcpServers"). ```jsonc { "servers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### Windsurf — `~/.codeium/windsurf/mcp_config.json` Open via Cascade → hammer icon → Configure. ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### Cline — `cline_mcp_settings.json` Open via the Cline sidebar → MCP Servers → Edit. ```json { "mcpServers": { "tavily": { "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### Continue — `~/.continue/config.json` Continue uses modelContextProtocolServers with a transport block. ```json { "experimental": { "modelContextProtocolServers": [ { "transport": { "type": "stdio", "command": "npx", "args": [ "-y", "tavily-mcp" ], "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } ] } } ``` ### Codex CLI — `~/.codex/config.toml` Codex uses TOML. Each server is a [mcp_servers.] subtable. ```shell # ~/.codex/config.toml [mcp_servers.tavily] command = "npx" args = [ "-y", "tavily-mcp", ] env = { TAVILY_API_KEY = "${TAVILY_API_KEY}" } ``` ### Zed — `~/.config/zed/settings.json` Zed calls them "context_servers". Settings live-reload on save. ```jsonc { "context_servers": { "tavily": { "command": { "path": "npx", "args": [ "-y", "tavily-mcp" ] }, "env": { "TAVILY_API_KEY": "${TAVILY_API_KEY}" } } } } ``` ### ChatGPT — `ChatGPT → Apps directory` Tavily doesn't ship a hosted HTTPS endpoint today. ChatGPT supports remote MCP servers only — to use this server in ChatGPT you'll need to deploy it to a public HTTPS URL first (e.g. via Cloudflare Workers or Vercel) or wait for an official remote build. ```none ``` ## At a glance - **Maintainer:** Tavily - **Transport:** stdio, Streamable HTTP - **Auth model:** API key - **Required secrets:** TAVILY_API_KEY - **Supported clients:** Claude, Cursor, VS Code, Windsurf, ChatGPT, Any MCP-compatible client - **License:** MIT - **Language:** TypeScript - **Latest version:** latest - **Last verified:** 2026-05-06 - **GitHub stars:** 1,944 (fetched 2026-05-11T09:29:10.957Z) - **Score:** 90/100 (rubric 2026-04 — see https://top-mcps.com/about/methodology) - **Source:** https://github.com/tavily-ai/tavily-mcp ## Tools & permissions | Tool | Description | Args | Side effects | |------|-------------|------|--------------| | `tavily-search` | Run a web search and return ranked results with snippets. | `query: string, topic?: string, max_results?: number` | network | | `tavily-extract` | Fetch and clean the body of one or more URLs. | `urls: string[]` | network | | `tavily-news` | Search news sources only, with publication-date metadata. | `query: string, days?: number` | network | ## Security & scope - **Access scope:** network - **Sandbox:** Calls the Tavily API over HTTPS with the key in TAVILY_API_KEY. The MCP itself only reads from Tavily — no write capability, no local filesystem reach. Rate limits and per-call cost are the main operational concerns. - **Gotchas:** - Per-call cost rises with `include_raw_content: true` — set a budget at the API-key level before unleashing an autonomous loop. - Tavily caches some searches; if you need real-time freshness for time-sensitive prompts, pass `include_raw_content: true` or pair with Brave Search for news-burst coverage. ## Quick answer **What it does.** Provides AI-optimized web search via the Tavily API, including standard search, news search, and combined search-and-extract that returns cleaned page bodies as Markdown. **Best for:** - Research-grade web search - News monitoring - Citation-ready summaries - Search + extract in one call - Agent-driven RAG workflows **Not for:** - Raw SERP introspection - High-volume free-tier needs - Search inside a private corpus (use a vector store instead) ## Recipes ### Citation-ready research summary ``` Using Tavily, search for the latest reporting on "Model Context Protocol adoption" with topic=news, max_results=8. Then call tavily-extract on the top 3 results and produce a 200-word summary with inline citations. ``` _Tested with: Claude Desktop, Cursor_ ## Description Tavily is a search API purpose-built for AI agents — every result is pre-cleaned, summarised, and shipped with source citations. The Tavily MCP exposes web search, news search, and a single-call "search and extract" path that returns Markdown bodies alongside the ranked list. Strong default for research-heavy agent workflows. ## Why it matters Brave Search returns raw SERP results; Tavily pre-cleans them for LLM consumption. For grounded research and citation-heavy answers, the difference is substantial — fewer follow-up Fetch calls, fewer hallucinated quotes. ## Key features - AI-optimized search results - News-only search mode - Search-and-extract single-call flow - Source citations in every result - Domain include/exclude filters ## FAQ ### How is Tavily different from Brave Search? Brave returns SERP-style raw results; Tavily returns LLM-cleaned summaries with source citations baked in. Brave wins for unfiltered research; Tavily wins when you want the agent to produce a citation-ready answer in one tool call. ### Does it support news-only search? Yes. Pass `topic: "news"` and Tavily restricts results to news sources, with publication-date metadata on each. Useful for monitoring, alerting, or "what changed in the last 24 hours" workflows. ### Can I get the actual page content alongside search results? Yes — Tavily's "search and extract" path runs a search and immediately fetches and cleans the top results in one round trip. Saves the agent a second Fetch call per source. ### What are the rate limits? The free tier covers 1,000 searches/month; paid plans scale up to 100k+. Set a daily budget on the API key for any agent workflow that loops, since per-call cost adds up faster with auto-extract enabled. ### Does Tavily work with ChatGPT through Developer Mode? Yes. Tavily ships a remote MCP endpoint that ChatGPT can connect to via Developer Mode. The same API key works across stdio and remote transports. ## Changelog - **2026-05-06** — Refreshed install snippets and fact sheet; verified for 2026. - **2025-03-15** — Initial directory listing.