Claude now has its own built-in browser in Cowork. When your task involves a website, a browser opens in Cowork's side panel, and Claude navigates, fills forms, and finishes the job.

There's nothing to install. Claude's browser is built into the desktop app and stays separate from your own browser and logins. Rolling out over the next week on the desktop app for all paid plans:

If you prefer to work in your own browser where you're already signed in, Claude in Chrome is now generally available on all paid plans. It remains your default if you already use it:

Wait it didn’t have this already? Wow I guess I was so spoiled by ChatGPT that I figured this was standard by now. Still a good improvement though.

All these small changes really streamline Claude’s Cowork workflow

When will it be able to put in passwords for us without knowing them?

CLAUDE MAKES SEEING BELIEVING 'When your task involves a website, a browser opens in Cowork's side panel, and Claude navigates, fills forms, and finishes the job.'

I’m so confused, I thought this already existed?

I sure wish @AnthropicAI spent more time actually fixing their models than always releasing more features for the model to mess up in…

We just want the long_conversation_reminder gone, man!

This is Great! The next killer feature set we need in Cowork (and code) is greater ability to edit docs in the app. Preview pane edits, annotations, batched edits, all that stuff. Just being able to type edits into a .md file would be awesome.

But why would I use that if it’s going to take usage away? Asking something about my projector trying to get information on it shouldn’t cost me usage. I hope that is a thing that eventually gets implemented. The thing that becomes unlimited

does the built-in browser share cookies with Chrome or is it a clean session

Great but why do you keep avoiding making Opus 5 usable?

This was already a thing?

thought that was a thing like 2 years ago now?

You banned my account and never refunded me my money. A whole 100$. You are useless

We need a Firefox browser plugin and not Google Chrome. Firefox is Open Source, while Google is CIA. Please add this extension to Firefox.

a form fill still has no schema and no idempotency key, so a retry can submit the same page twice. last silent 200 i had was a renamed column the MCP server kept serving. you doing a read-back of the object after the submit, or trusting the toast?

Netscape in 94 tried to build the entire web into a desktop OS, and we all know how that ended when the browser became the OS. Claude turning into the agent for every tab is just Netscape's dream finally working without the memory leaks.

wait what?? nothing

Good. The wrapper economy around AI web scraping was getting tedious. If this works natively, half those startups are finished.

I've been amazed at how good browser use is now. Might not need an MCP server for an app unless you have bulk change operations or complex UI steps

Claude Desktop (linux) needs an urgent fix. Thanks. Lost an entire thread, literally vanished. Opus was able to find some trace of the task that was interrupted but everything was lost. Don't waste time talking about humanity nobody cares. Fix your stuff.

sidebar browser is the right shape — keeps the model from grabbing whatever happens to be on the user's tab. bigger tell is whether it ships with site allowlists or lets Cowork roam free. that decides whether this is a productivity toy or an actual agent surface.

The real test is how it handles CAPTCHAs and anti-bot protections. If it can bypass those natively in the side panel, this is an absolute game changer for agent workflows.

Everyone is impressed by the panel, but the panel is the part that does not matter. It's live and then it's gone. What was actually typed into that form? I need to be able to check what actually happened.

A big step towards background computer use!

Would love the mobile app to work. It's why I use ChatGPT Work as my daily driver.

WebMCP support?

Not bad at all. A built in browser in Cowork isolated and with nothing to install is genuinely useful for web tasks without touching your real browser.

At this point, I just feel like they are just playing catch up with ChatGPT.

Fix the ssh and remote issues already. Even your feedback doesn’t work