The grading is the design worth having: interactive elements when you are about to click something, everything visible when you need context, text alone when you are reading. Each costs a different amount of context, and choosing the wrong one is the usual way a browser tool turns expensive. execute-code is the fallback for whatever the snapshots do not reach. One thing to know before relying on it: get-full-dom's own description marks it deprecated in favour of get-context, and no tool by that name appears beside it.
A Playwright browser server with one initialisation tool and several differently-scoped ways of reading the current page, plus arbitrary code execution against it.
- init-browser, opening a browser at a URL
- get-interactive-snapshot, focused on buttons, links and inputs, with an annotated screenshot for driving the interface
- get-full-snapshot, everything visible on the page — text, images, forms — for understanding full context
- get-text-snapshot, text only: headings, paragraphs and lists, for reading and content extraction
- get-screenshot of the current page
- execute-code, running custom Playwright JavaScript against the page
Nothing — no account, no key. It drives a browser on the machine it runs on.
One command — npx -y playwright-mcp