There are two kinds of maintenance in web scraping: fixing selectors when a site is redesigned, and getting past whatever stands between your request and the page. A Python library called Scrapling, trending on GitHub on Wednesday with about 1,661 stars in a day, is an attempt to sell both as one product.
The first half is the parser. Scrapling learns a selection — fetch a page, call css('.product', auto_save=True) — and then, after the site's markup changes, re-query with adaptive=True and it relocates the elements by similarity instead of failing. For anyone who has maintained a scraper across a redesign, that is the tedious part of the job done by a library rather than a person.
The second half is stealth. StealthyFetcher spoofs browser fingerprints, the HTTP layer can impersonate a Chrome TLS fingerprint and speak HTTP/3, and there is optional DNS-over-HTTPS so requests sent through proxies do not leak DNS. The README claims it clears Cloudflare's Turnstile and interstitial challenges. There is a proxy rotator with cyclic or custom strategies across session types, domain blocking, and remote browsers over CDP.
Around that sits a real crawling framework: Scrapy-like spiders with start_urls and async parse callbacks, concurrency limits, per-domain throttling that doubles a delay when a site starts blocking and relaxes again when it stops, pause/resume checkpoints so Ctrl+C is safe, streaming results, a development mode that caches responses so you can iterate on parsing without re-hitting the target, and templates including a ShopifySpider that walks a store's JSON catalogue.
Two details say where the project thinks scraping is going. First, robots.txt compliance is a flag (robots_txt_obey) rather than a default. Second, the library now ships an MCP server and an "agent skill" so coding assistants write correct Scrapling code instead of hallucinating the API — extraction is being packaged as a tool for language models, not a library for humans.
The business model is visible on the README: the platinum sponsor block is proxy vendors, and one sponsor sells antibot tokens for Akamai, DataDome, Kasada and Incapsula. The bypass is free software; the IP reputation is the paid product. That is the real cost curve of scraping in 2026 — parsing speed was solved years ago (Scrapling benchmarks its parser at 1.99 ms against 1,562 ms for BeautifulSoup with lxml, on its own methodology), and what remains expensive is looking like a person.
One caveat is worth stating plainly: getting past a bot check can breach a site's terms even where the scraping itself is lawful, and a library cannot make that decision for you. The project's own disclaimer points users at local law and terms of service.
The larger point is that the anti-bot bypass has been commoditised. What used to be a specialist service is now a paragraph of Python with a sponsor list. Whatever moat a site's bot defence provided now costs roughly the price of residential bandwidth.
