If you’re busy using LLMs to browse the web, have a look
Gemini already has one, just call the /browser skill in any prompt, but alternatives are nice. We developed a custom one to consult databases (the ones in the wild don’t behave well enough)
Not quite the same. Browsing the web is simple, but literally everything else uses DOM scraping for it. Which is very heavy on token usage. This is the first AOM driven browser MCP. For perspective, sites like Amazon and Wikipedia take up 100k+ tokens just loading the page into context. using the AOM, it’s a fraction of that. So no, it doesnt already have one, it has the dumb tool built on selenium, which uses the DOM, which is extremely token heavy when dynamic and extremely flaky when scripted.
Ooooh, you mean pure navigation, not for developing. Yeah AOM seems better for that, now the issue it that every page needs to implement them properly (not a bad thing).
AOM is mandatory across the web for accessibility. If a website isnt compliant to the standard, they can get into serious trouble for it, because it’s what screen readers use to let disabled people navigate.
That’s totally true, but there are also poorly implemented ones, for a fact I know a good bunch of them, they are mostly internal tools.