puppeteer_connect_active_tab is the distinguishing tool: attaching to a Chrome that is already running means the cookies, extensions and logged-in state are there, so a flow behind an SSO wall does not need a scripted sign-in that will break the next time the login page changes. The rest is the familiar set — navigate, click, fill, select, hover, screenshot — with puppeteer_evaluate as the escape hatch. Everything addresses elements by CSS selector, so a page that only distinguishes controls visually will still be awkward.
A Puppeteer driver that can either work on its own page or connect to a running Chrome over its remote debugging port.
- A connection to an existing Chrome instance over its remote debugging port, optionally aimed at a target URL
- Page interaction by CSS selector: click an element, fill an input, choose an option in a select, hover
- Screenshots of the whole page or of a single element, at a width and height you set
- puppeteer_evaluate, which runs JavaScript in the page console when a selector is not enough
Chrome on the machine, started with remote debugging enabled if you want to attach to a session you are already signed into. MIT licensed.
One command — npx -y puppeteer-mcp-server