I've done this, but I don't remember if I did it with goose or manual screenshots and the Claude chat interface. It worked pretty well
Whenever I vibecode UIs it feels like the agent is flying blind because it can't see what's being rendered.
Have any of you tried integrating a browser-automation MCP like this one? Seems like this could really help the agent QA its work.
https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer