krishnashakula/browsewright — reverse-engineered prompt

Build me a small Python tool called Browsewright that lets me give it any web page and a plain English goal, then it figures out the page on its own and returns a useful answer or clean JSON. I want it to drive a real Chrome browser when needed, so it can actually click around, read labels, fill out forms, submit them, and pull back structured results instead of raw HTML. If there is an easier cheaper path, like a public API, RSS feed, or archive, try that first before opening the browser.

It should work from the command line and also as a Python library, with simple examples. Please include a few ready to use tasks like lead enrichment, competitor watching, price or stock tracking, brand monitoring, and form filling from a profile file. It should run locally with my own Anthropic key, support headless or visible browser mode, and be able to expose a simple MCP tool so another LLM can call it. Add basic tests and docs, and look up current docs online if you need to.