Summary
Selectors decide what the model reads, so a long page costs a paragraph.
The selectors parameter is the whole argument for it: without one the page comes back whole, with one only the matched elements do. That makes it a fit for pages whose structure you already know, since the tool takes selectors rather than working out which part mattered.
What it is
A single-tool page extractor: a url, plus optional CSS selectors naming the elements worth returning.
What you get
- browse_webpage returning the content of a page at a url
- Selectors narrowing that to matched elements, so a table or an article body arrives without the navigation and footer wrapped around it
Requirements
Nothing to supply — no account, no key; a URL the machine can reach, and the selectors for the part you want. MIT licensed.
Setup effort
One command — uvx web-browser-mcp-server