Scavio icon

Scavio

Real-time structured data from Google, Amazon, Walmart, eBay, Target, Home Depot, Costco, YouTube, Reddit, TikTok, Instagram, X, LinkedIn, Facebook, Pinterest, Threads, Kuaishou, Douyin, Weibo, Twitch, GitHub, Etsy, AliExpress, Zillow, Redfin, Realtor.com, Rightmove, Booking.com, Airbnb, Tripadvisor, Yelp, Indeed, Glassdoor, the App Store, Google Play, SEC EDGAR, Companies House, Legistar, G2, Capterra, Trustpilot, ad libraries, and any URL

Actions227

Extract (Any URL) → Extract Page

AI-generated

Summary

Fetches a web page at the supplied URL and returns its content as raw HTML, extracted Markdown, or plain text. Choose Normal for a standard fetch, Advanced for JavaScript-rendered pages, or Ultra for harder targets; successful extractions consume 1 credit for Normal or Advanced and 2 for Ultra.

Inputs

  • URL (required) — HTTP or HTTPS page URL; bare hosts are upgraded to HTTPS. Loopback, private, link-local, and metadata hosts are rejected.
  • Output Format — Content shape: HTML returns the page as served, Markdown extracts the main content and removes boilerplate, and Plain Text flattens that Markdown. Defaults to Markdown; format does not affect price.
  • Mode — Fetch strategy: Normal uses a plain datacenter fetch, Advanced uses a headless browser for JavaScript-built pages, and Ultra uses a premium residential proxy. Defaults to Normal. Normal and Advanced cost 1 credit; Ultra costs 2. Failed extractions are not billed.
  • Request Options — Optional request controls: batch size and interval for throttling requests, whether to accept invalid SSL certificates, an HTTP proxy, and a response-header timeout in milliseconds. Defaults include 50 items per batch, a 1000 ms interval, SSL validation enabled, no proxy, and a 10000 ms timeout.

Output shape

A response containing the extracted page content.

The selected format determines the content representation. The source pack does not expose the runtime response fields or a more specific return shape.

Links

Discussion