Scrapling Parses 5,000 Elements in 1.99 Milliseconds. The Cloudflare Part Is the Harder Sell.

D4Vinci/Scrapling hit the GitHub weekly chart with about 8,700 stars in seven days, sitting at 99,671 total. It is a Python scraping framework, which is a crowded category, and it is not competing on being small. The pitch in the first line of the README is that it puts fast HTTP requests, a stealth browser, parsing and concurrent crawling into one library, which is a description of four libraries that usually have to be glued together.
The number people quote is a benchmark. Text extraction across 5,000 nested elements: Scrapling at 1.99 ms, Parsel/Scrapy at 2.06, raw lxml at 2.56, PyQuery at 23.98, Selectolax at 197.02, MechanicalSoup at 1,545.15, BeautifulSoup with lxml at 1,562.1, BeautifulSoup with html5lib at 3,412.73. Averages over a hundred runs, per the README.
Read that table twice and it says something narrower than it appears to. The top three are within 30% of each other, and all three are lxml-backed parsers with different wrappers. The 780x gap against BeautifulSoup is real, and it has been real for a decade. Scrapling lands within noise of lxml while doing far more than parse, which is a genuinely good outcome and a different claim.
The install has two steps and the first one is a trap
pip install scrapling
That installs the parsing engine and its dependencies. Nothing else. Import from scrapling.fetchers or scrapling.spiders and you get a ModuleNotFoundError, because the fetchers ship as an extra:
pip install "scrapling[fetchers]"
scrapling install
The second command downloads browsers and system dependencies. Skip it and the stealth fetchers have nothing to drive. Other extras: scrapling[ai] for the MCP server, scrapling[rag] for markdown output aimed at retrieval, scrapling[shell] for the interactive shell and the extract command, scrapling[all] for everything. Docker images exist at pyd4vinci/scrapling and on the GitHub registry.
The split is defensible. A parser has no business pulling in a browser. It still means every tutorial that starts with pip install scrapling produces a traceback on the second line, and the README spends its first warning on exactly that.

Two fetchers, two different jobs
The basic call is the one everybody writes:
from scrapling.fetchers import Fetcher
page = Fetcher.get('https://quotes.toscrape.com/')
quotes = page.css('.quote .text::text').getall()
FetcherSession(impersonate='chrome') keeps a session and impersonates a browser's TLS fingerprint, which matters for sites that fingerprint the handshake rather than the user agent string. It also supports HTTP/3.
The other fetcher is the one the project is really selling:
from scrapling.fetchers import StealthyFetcher
page = StealthyFetcher.fetch('https://nopecha.com/demo/cloudflare', solve_cloudflare=True)
StealthySession keeps a headless browser open, can solve Cloudflare Turnstile and interstitial checks, and supports record_requests=True plus session.network.search() to capture the background API calls a page makes. That last feature is the one that changes workflows. A lot of scraping work is not reading HTML at all, it is finding the JSON endpoint the page calls and calling it directly, and a tool that surfaces those requests saves an afternoon in dev tools.
There is also ProxyRotator for cycling proxies, DoH-based DNS leak protection through Cloudflare, blocking for roughly 3,500 known ad and tracker domains, and a cdp_url option for attaching to a remote browser.
The Spider API is Scrapy-shaped on purpose
from scrapling.spiders import Spider, Response
class QuotesSpider(Spider):
name = "quotes"
start_urls = ["https://quotes.toscrape.com/"]
concurrent_requests = 10
async def parse(self, response: Response):
for quote in response.css('.quote'):
yield {"text": quote.css('.text::text').get(),
"author": quote.css('.author::text').get()}
nxt = response.css('.next a')
if nxt:
yield response.follow(nxt[0].attrib['href'])
result = QuotesSpider().start()
result.items.to_json("quotes.json")
start_urls, async parse, yielded dicts, Request and Response objects: anyone who has written a Scrapy spider can read this without a doc tab. The framework adds per-domain throttling with AutoThrottle, robots_txt_obey, a LinkExtractor, blocked-request detection with retry, and a development mode that caches responses to disk so you can replay a crawl without hitting the site again.
crawldir is the feature that will save the most time in practice. Pass a directory and Ctrl+C persists pending requests and seen URLs; restart with the same directory and the crawl resumes. Long crawls fail, and a framework that assumes they will not is a framework you babysit.
Templates ship for the common shapes: CrawlSpider, SitemapSpider, XML and CSV feed spiders, and ShopifySpider, which reads a store's public collection endpoints and emits one item per variant. Export is built in: to_json, to_jsonl, to_csv, to_xml.
The selector object is the underrated part
Parsed pages come back as a Selector object rather than a list of nodes. It supports CSS, XPath, find_all, find_by_text, regex search, and navigation: .next, .parent, .below_elements, find_similar(). There is an element-tracking feature that relocates a target after a site redesign, which is the maintenance problem that kills long-lived scrapers. It will not always work, and when it does, it saves a rewrite.
The parser also runs standalone: Selector("...") with no network call, which is how you unit-test extraction logic.
The MCP server, and why a scraping library now ships one
scrapling[ai] adds an MCP server so an agent can call the fetchers directly, including screenshots, logged-in sessions and captured background APIs. The README notes prompt-injection protection, which is the right thing to worry about: a scraper that hands page text to a model is a channel for whatever the page wants the model to read.
That packaging choice says where the maintainers think the users are. A year ago, a scraping library was used by a data engineer inside a script. Now it is also a tool an agent picks up mid-task, and the interface it needs is a tool call with a schema.
The line in the README that does the real work
Buried near the end: "This library is provided for educational and research purposes only." Plus a request to respect terms of service and robots.txt.
That is a standard disclaimer and it does not do much legally. It does describe the actual situation accurately. The technical capability to get past a bot check and the legal permission to do so are unrelated, and the second one is getting worse. Sites that used to tolerate scraping are now suing, blocking at the edge, or licensing access. A framework that makes bypass trivial shifts the decision from "can I" to "should I", and it shifts it onto whoever runs the script.
The benchmark table will get the stars. The install trap will get the issues. The part that determines whether this repository is useful in two years is whether the maintainers keep the stealth layer current, because that is a treadmill and parsing is not.
Related articles
Chinese text in video always turns to mush. Dreamina handed the typesetting to code
Dreamina launched Motion Studio on November 10. It no longer generates frames; the model writes code that lays out text, shapes and pacing. Chinese titles stop blurring, but whether the content is right is still on you.
A community spreadsheet of taps that turns off the ads inside Chinese Android builds
A no-root tool that switches off the ads Chinese phone makers ship inside the system UI. Its method is a hand-written tap script per phone model, maintained by volunteers every time a ROM moves. It works, and the maintenance is the point.
GPT Image 2.6 held the token price and raised the bill
GPT Image 2.6 kept every token rate from 2.5 and still costs more per image, because more of the work now happens in output tokens. Eleven days into the ramp, the pricing structure says more than the leaderboard does.
Someone rebuilt IDM from the outside and wrote down what they did not touch
A Rust reimplementation of Internet Download Manager claims 5x the speed and 86 languages. The more interesting document is the paragraph explaining how it was built without copying anything.