Summary of an article published on The Web Scraping Club on 2026-10-06 (free to read). Read the full article
A local browser comes with logins, saved sessions, a clean residential IP and real hardware. An agent running on a Linux VM in a data center has none of that, and its IP raises red flags from the first request. For an agent that has to act as a logged-in user, that gap is where things break.
For this Tool of the Week we used Browserless to build an agent that logs in to an Italian grocery site and adds a product to the cart. We worked with Playwright over CDP, the stealth route, the built-in Italian residential proxy and the native CAPTCHA solver. We tried two targets: Esselunga, behind F5 BIG-IP, and Tigros, behind Cloudflare with reCAPTCHA Enterprise. Then we tested Authenticated Profiles and the Session API, to see whether a new browser could skip the login. We also looked at how Browserless bills units.
What’s inside
- How the stealth and plain Chromium endpoints compared on the first Esselunga request, and which anti-bot signal showed up.
- What happened with the second factor on Esselunga, and why our hypothesis about the Linux identity could not be proven.
- Why we used reCAPTCHA Enterprise’s invisible, score-based version on Tigros and how the native solver behaved against the page’s own token.
- Profile versus persisted session: a copy by value against a browser identity that keeps changing on disk, and what changes when the exit IP changes.
In the full article: The Python and Playwright code, the connection parameters, the CAPTCHA solve results per run, the profile and Session API calls, the unit costs and the cart test on Tigros.