Scrape a single URL and return its content as markdown, HTML, or links. Use this to extract content from any web page.

Start an async crawl of a website. Returns a job ID that can be polled with crw_check_crawl_status.

Check the status of an async crawl job and retrieve results.

Discover URLs on a website by crawling and/or reading its sitemap.

Search the web and return relevant results with titles, URLs, and descriptions/snippets. Backed by a SearXNG sidecar in embedded mode (no API key needed), or by the configured remote API in proxy mode (uses CRW_API_KEY).

Return shape: { "success": true, "data": { "results": [{ "url", "title", "description", "snippet", "position", "score" }, ...] } }. When sources is set, data.results is instead an object grouped by source ({ "web": [...], "news": [...], "images": [...] }). The snippet field is an alias of description — both carry the same body text so downstream LLM pipelines that ask for either get a match.

Example: crw_search(query="renewable energy trends 2024", limit=3) returns the top 3 web results with title/url/snippet.

Errors: returns search_disabled when no SearXNG backend is configured, or target_unreachable / timeout (naming the configured host) when the backend can't be reached.

Parse an uploaded document (PDF) and return its content as markdown. Pass the file bytes base64-encoded in contentBase64. Use this for local PDFs that have no URL. Scanned/image-only PDFs have no text layer (no OCR) and return a warning with empty markdown.