Tavily MCP — Web Search + Extraction

Search the web, map site URLs, extract structured content, or crawl with breadth/depth controls. Use when you need current information plus reliable page text extraction.

Codex CLI Notes

• •Stdio transport via $HOME/mcp/mcp-zotero-matlab-tools/bin/tavily-mcp.sh (sets Volta/XDG caches under $HOME/mcp/mcp-zotero-matlab-tools). Call tools by name; prefer search first, then extract/crawl selectively.

Tools and When to Call

• •tavily_search(query, max_results?, topic?, search_depth?) — Web/news search with recency options.
• •tavily_map(url, max_depth?, max_breadth?) — Discover URLs on a site.
• •tavily_extract(urls, format?, extract_depth?) — Extract clean page content (markdown/text).
• •tavily_crawl(url, max_depth?, limit?, select_paths?, exclude_paths?) — Multi‑page crawl (summaries per page).

Examples (codex-cli)

• •tavily_search {"query":"OpenAI o3 mini system prompt", "max_results":5}
• •tavily_map {"url":"https://example.com/docs"} → pick URLs → tavily_extract for full content

Guidelines

• •Start with tavily_search; keep extraction targeted to avoid noise.
• •Use select_paths/exclude_paths and low depth to focus crawls.

Security

• •Keep API keys in environment (e.g., TAVILY_API_KEY). Respect robots.txt and site ToS.