Summary
A browser the assistant can actually operate.
Because it drives a real Chromium rather than fetching HTML, pages that only exist after JavaScript runs are reachable — and the console log comes back with them, which is what turns the page looks wrong into something you can read.
What it is
A Puppeteer-backed browser automation server. It opens pages in a real Chromium, acts on them through CSS selectors, and exposes the browser console log and the screenshots it captures as MCP resources.
What you get
- Any URL opened in the browser
- Screenshots of the whole page or of one element, at the width and height you set
- Clicks, hovers, form fills and select-menu choices, addressed by CSS selector
- JavaScript executed in the page console
- Browser console output readable at console://logs
- Each screenshot kept at screenshot://<name> for later reference
Requirements
Node with npx, or Docker. The Docker image runs headless Chromium; the npx path opens a visible browser window.
Setup effort
One command — docker run -i --rm --init -e DOCKER_CONTAINER=true mcp/puppeteer