Summary
A page can be operated without anyone writing a selector.
browser_snapshot returns the page as an accessibility tree with a reference on each element, and every interaction call takes one of those references — so the agent aims at what the page exposes rather than at a CSS path that changes with the next deploy. When something goes wrong, the console logs and a screenshot are both one call away.
What it is
Twelve tools for a browser session: navigation and history, an accessibility snapshot that gives every element a reference, interaction aimed at those references, and two ways to see what happened — console logs and a screenshot.
What you get
- Navigation to a URL, and back and forward through history
- An accessibility snapshot of the page, which is where the element references every interaction call takes come from
- Clicks, hovers, typing with an optional submit, dropdown selection and key presses, each aimed at a named element and its reference
- A fixed wait in seconds for pages that need one
- The browser's console logs, and a screenshot of the current page
Requirements
Npx on your PATH. Interaction calls take an element and its reference, so browser_snapshot has to run before the first click.
Setup effort
One command — npx -y @browsermcp/mcp