The moment Claude says it can't check

The moment it says "I can't check"

Ask Claude to check today's price on a page you have open, and sometimes you get an answer you don't expect: it can't. Its knowledge stops at a training cutoff, and browsing a live page isn't something a plain conversation can do. Most people shrug, open a new tab, and do it themselves.

That's the wrong response, because there's a fix, and it isn't waiting for a smarter model.

What a connector actually wires up

A connector is a wire from Claude into something outside the conversation entirely: a live search engine, a real browser, the actual content on a website, the tabs already open on your screen. Once that wire exists, Claude stops describing what it would do and starts doing it.

Five of these are worth naming, because each one answers a genuinely different question.

Perplexity gives Claude live web search, for the question that changed after training ended: a competitor's announcement from last week, a price that moved yesterday.

Playwright hands Claude a real browser it can click through, filling in a form and testing a signup flow the way a person would. It's the connector that turns "describe how the bug happens" into "here, I reproduced it."

Firecrawl reads a site properly and pulls out what actually matters, instead of a search-engine snippet of it.

Glyph wires Claude to image, video and audio generation, for teams whose output isn't text.

Chrome gives Claude the tabs you already have open, so it works on what's already in front of you instead of you pasting it in.

The Sunday night that doesn't need to stay manual

Picture a shop owner who spends every Sunday night opening ten competitor pages by hand, checking who dropped a price before the week's orders go out. It's the kind of task that eats an evening and produces nothing you'd call interesting: open a tab, scan a number, close it, open the next one.

That's exactly the shape of task a connector removes. Wire Firecrawl or Playwright to those ten pages once, and Claude comes back with the numbers instead of you clicking through them yourself. The task doesn't get smarter. It just stops being yours to do by hand.

Don't install the whole shelf

The instinct, looking at a list of five, is to install all five. I'd resist that. A connector isn't a feature you flip on and forget. It's a capability Claude now has permission to reach for, and every one you add is something worth understanding before you trust it with a task that matters. Five idle connectors don't make Claude five times more capable. They make your setup five times harder to reason about when something behaves oddly.

Start from the task, not the list

Name the task that actually repeats in your week first. The competitor prices you keep checking by hand. The research you keep copying out of ten open tabs. The question Claude keeps answering with "as of my training." Then find the one connector, maybe two, built for exactly that. Everything else can wait for a task that needs it.

Watch the first few runs closely

A connector that can act on your behalf, clicking, filling in forms, browsing while you look elsewhere, deserves the same attention you'd give a new hire's first week. Check what it actually did before assuming it did the right thing, and keep checking for longer than feels necessary.