tavily-ai Tavily Search
Find pages and current information when you do not already have a URL.
When
- The user needs information on a topic and has not supplied a URL
- First step: search → extract → map → crawl → research
Do not use this skill when the user already has URLs (extract), needs every page in a site section (crawl), needs to locate pages on a known site (map), or wants a cited multi-source report (research).
How
- Write short search queries, not prompts — keep each under 400 characters. Split multi-part questions into separate queries.
- Apply domain, topic, or date constraints when the user specified them.
- Prefer primary sources for factual or technical claims. Cross-check consequential facts.
- Cite source URLs. Synthesize; do not paste raw result dumps. Do not request full page HTML for every hit; if snippets are thin, extract the best URLs instead.
If snippets are not enough, extract the best URLs. If the user needs a full cited briefing, switch to research.
Execute
If tavily_search is callable (the name may be prefixed), call it. Do not install the CLI and do not run tvly for this. In a coding-agent terminal that is already using tvly, pass --client-name "cursor plugin" and take flags from tvly search --help. Never ask for an API key in chat.