Scrapling hits 69,000 GitHub stars on adaptive selectors

Scrapling has crossed 69,000 GitHub stars, per the project's GitHub page. For anyone running agents against the web, that number tracks real cost. A brittle selector means a silent failure or a re-scrape, and either one burns time and tokens.

Built by D4Vinci, Scrapling learns each element's signature instead of relying on fixed CSS or XPath rules. When a page's structure shifts, it relocates the element by that signature and keeps going instead of breaking outright.

It also gets through anti-bot walls like Cloudflare Turnstile, using browser-level rendering and fingerprint rotation so the scraping step doesn't stall before it starts.

The part worth building around is the MCP server. Instead of dumping raw HTML into a model's context, Scrapling's MCP server extracts the targeted content first and hands the model a narrowed result.

The pitch is token math

I haven't run Scrapling against my own agents yet, but the pitch is direct. Front your agent's web steps with MCP extraction so the model never sees raw HTML at all. That's a token cost cut, not a speed trick. 69,000 stars says a lot of people are already paying the raw-HTML tax and looking for a way out.

The GitHub page doesn't say how the signature-matching holds up against a full site redesign, only against structural drift. Worth testing before you trust it on a page that gets rebuilt from scratch.