Two tools. A Cloudflare wall between them. They ran the same jobs side by side, and here's what each one actually did.
I've used Playwright for a long time, so to keep it fair I installed both tools from scratch, gave them the same two tasks, and watched what each one did.
Setup
I know my way around Playwright. Install it, launch headless Chromium and you're all set, in general. But as soon as your network isn't perfect, it starts throwing ERR_QUIC_PROTOCOL_ERROR and won't work until you disable QUIC, ECH, and DNS-HTTPS-SVCB with Chromium flags. Fussy tool.
BrowserAct is a CLI, and getting going was a single line:
uv tool install browser-act-cli --python 3.12
That's all, installed and ready to use.
Round 1: an open JavaScript page
Warm up: scrape all ten quotes, text, author, and tags, from quotes.toscrape.com/js, a JavaScript-built page.
Playwright consumed it. Around twenty lines, a couple of seconds, identical outcome every run:
BrowserAct also pulled it off with just one command and no parsing code:
browser-act stealth-extract https://quotes.toscrape.com/js/ --content-type markdown
Clean markdown, all ten quotes, and I didn't write a single selector, no scraping logic, nothing to re-maintain when the page shifts. One line in, structured data out.
Round 2: the Cloudflare wall
Same URL for both: scrapingcourse.com/cloudflare-challenge.
Headless Playwright crashed into the wall and came to a halt. Cloudflare's "Just a moment" screen, the little spinner, and then nope. Nothing behind it, not even after we gave it the old wait-around:
BrowserAct walked straight through. A single command, about 31 seconds, and it handed me the page's own line:
Here's what that gap really costs you. To pull that off in Playwright yourself, you're setting up a stealth browser, renting residential proxies, and integrating a captcha solver, three moving parts you now own forever. BrowserAct folds all of that into one command and just serves you the page.
The human handoff
The part I didn't see coming is remote-assist. It's for the steps a bot shouldn't be doing anyway, a login, a 2FA code, one of those "tap every crosswalk" grids.
I tested it with a throwaway login. A single command gave me a live URL and the agent went quiet. I opened the link, typed the password myself, handed control back, and it picked up right where it left off, still logged in:
This is the hurdle where most scrapers cease to exist, and it simply finds a way around it.
Parallel work and account isolation
BrowserAct was built for parallel jobs and isolated accounts, so I put that to the test.
Two sessions, one browser, one on quotes.toscrape.com and the other on books.toscrape.com, both live at once. They ran cleanly and independently. Then I got curious: I dropped a cookie in the first session and went looking for it in the second, and there it was:
Sessions in one browser share a cookie jar, great if you're driving one account across a dozen tabs. If you want fully separate accounts, you spin up a browser with its own jar and fingerprint. I liked that it was clear about exactly where that line sits.
Skills
There are also get-skills, pre-built playbooks for the fiddly parts, logins, multi-browser setups, debugging. The whole library is open: github.com/browser-act/skills. I used it to get the CLI working; I didn't build my own.
The verdict
Both tools, the same jobs, and BrowserAct was the clear winner.
Playwright's good on open, JavaScript-heavy pages, fast and free. But that's the easy half of the web. The second a site actually defends itself (which pretty much covers most meaningfully important websites) with Cloudflare, Turnstile, a captcha, Playwright is done.
It sat at that wall and never moved. BrowserAct went through the same wall, and it does it while folding the stealth browser, the proxies, the captcha solving, and the human handoff into commands you can type from memory. That's an entire stack you'd otherwise build, run, and babysit yourself.
Most of what I scrape these days sits behind some kind of bot wall. If yours does too, BrowserAct is the one I'd reach for. Check it out at browseract.com.
Try BrowserAct yourself: github.com/browser-act/skills
All features are free — the only paid add-on is built-in proxies for IP rotation. Star the repo, and you'll earn credits toward your account.
Top comments (0)