The tool surface is the familiar navigate/click/fill/screenshot set; what you are choosing here is the escape hatch. Anything Puppeteer accepts at launch — a different browser, a fixed viewport, extra flags — goes in one environment variable, which is the difference between working around a site and forking the server.
A Puppeteer-backed browser automation server whose distinguishing feature is that the browser launch options are yours to set. A JSON string in `PUPPETEER_ARGS` is passed straight to Puppeteer, so you can switch the engine or fix the viewport without touching the server code.
- Navigation to any URL — `puppeteer_navigate`
- Screenshots of the whole page or one element, with width and height you choose (defaults 800 by 600) — `puppeteer_screenshot`
- Clicking and hovering elements by CSS selector — `puppeteer_click`, `puppeteer_hover`
- Form input and dropdown selection — `puppeteer_fill`, `puppeteer_select`
- Arbitrary JavaScript executed in the page — `puppeteer_evaluate`
- Two resources to read from: browser console output at `console://logs`, and each capture at `screenshot://<name>`
No account and no key. Launched straight from GitHub with `npx -y github:afshawnlotfi/mcp-configurable-puppeteer`, optionally pinned to a branch, tag or commit. `PUPPETEER_ARGS` takes a JSON string of Puppeteer launch options — the README shows `{"browser": "firefox"}` and a `defaultViewport` of 1280 by 800.
One command — npx -y github:afshawnlotfi/mcp-configurable-puppeteer