tavily-ai Tavily Extract
Pull clean page content from URLs you already have.
When
- The user provided URLs, or search already found the right pages
- Step 2: search → extract → map → crawl → research
Do not use this skill to discover pages (search or map) or to bulk-collect a whole site section (crawl).
How
- Extract only the URLs you need, at most 20 per call. Batch larger lists.
- Prefer query-focused extraction on long docs instead of entire pages.
- Retry with a deeper extract when the page is JavaScript-heavy, protected, or table-heavy and the first pass is empty or thin.
- Skip extract when search results already include the content you need.
- Cite URLs. Summarize for the user; do not dump full page text unless they asked for it.
If many pages on one site are required, crawl. If you still need to find the right path on a large site, map first.
Execute
If tavily_extract 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 extract --help. Never ask for an API key in chat.