Claude in Chrome: Getting Claude Into the Tools You Already Use
A plain-English look at Claude in Chrome: how the browser extension's permission model works, and four concrete workflows worth trying in a small business this month.
Last week we covered Cowork, the part of Claude that hands off an outcome and works on it in the background. This week is about the other half of "getting Claude into your actual work": Claude in Chrome, the browser extension that lets Claude act inside the tools you already have open, logged in as you.
What it actually does
Claude in Chrome adds a side panel to your browser. Open it on any page and Claude can read what's on your screen, and (with your permission) click buttons, fill in forms, switch between tabs, and carry out multi-step actions across the sites you're already signed into. The practical difference from chatting with Claude is that it isn't just describing what to do, it's doing it, inside Gmail, Salesforce, a dashboard, or wherever the task actually lives.
It's had a genuinely new development since we wrote about Cowork: as of mid-August 2026, the Chrome side panel stopped being its own separate, disconnected conversation and now runs as a full Cowork session on Max and Team plans (rolling out to Pro). That means a task you start in the browser saves to your history and picks up where you left off on desktop or mobile, and skills and connectors you've already set up work in the browser too. The two features from this series are converging rather than staying separate lanes.
The benefit: you're not maintaining two different "modes" of Claude: one for chat, one for background work, one for the browser. It's one assistant with context that follows the task, not the device.
The safety model, in plain English
The obvious question with anything that can click and type on your behalf is "what stops it from doing something I didn't want." Anthropic's answer is a permission system, laid out in its Chrome safety guide and permissions guide: Claude asks before acting on a site for the first time, and you choose between allowing a single action or trusting it on that site going forward. Certain things stay gated no matter what you've previously allowed (downloading a file, entering sensitive information, granting an authorisation) and some things are off-limits entirely, including financial transactions, entering payment card details, permanently deleting things, and bypassing a captcha. If you run a team, admins can set allowlists and blocklists for which sites the extension can touch at all, account-wide.
The benefit: the guardrails aren't a settings page you have to configure correctly to be safe. The risky categories are blocked by default, and everyday browsing on a trusted site is where you get to decide how much friction you want.
Four workflows worth trying
A handful of examples from Anthropic's own use-case library map well onto ordinary small-business admin, and they're a reasonable place to start rather than inventing your own from scratch.
Pulling numbers into a summary. Point Claude at an analytics dashboard and ask for the current month's key figures written up in plain language: it reads the dashboard and compiles the summary instead of you screenshotting and retyping numbers.
Logging a sales call into the CRM. Claude can match a calendar meeting to the right CRM contact and draft the activity log entry, which you approve before it's created. Useful for the admin that usually happens (or doesn't) right after a call ends.
Comparing options across open tabs. Open a handful of product or vendor pages, tell Claude which specs matter, and it can pull the details into a comparison table with a recommendation. Handy for supplier or software decisions where you'd otherwise be tab-switching and copy-pasting into a spreadsheet by hand.
Clearing out inbox clutter. Claude can scan for promotional and marketing email and group it for a bulk decision, rather than you deleting one at a time.
The benefit: none of these are dramatic on their own, but they're exactly the small, repeated tasks that quietly cost a business owner an hour here and there every week, and each one is something Claude can do with you watching and approving, not something you have to hand over blind.
Start with something low-stakes
If you're trying this for the first time, don't start with anything transactional or tied to money. A research or summarising task (the dashboard pull, the product comparison) is the right first move: low risk if it gets something wrong, and it's the fastest way to get a feel for the permission prompts before you decide which sites are worth "always allow." Once that feels comfortable, moving on to something like CRM logging, where Claude is writing into a system of record, makes a lot more sense.
Bringing it together
Cowork and Claude in Chrome are starting to look like two doors into the same room rather than two different tools: one for work that runs in the background, one for work that happens on a page you're already looking at, and now, increasingly, the same conversation either way. The extension is worth ten minutes of setup even if all you do with it at first is a couple of research tasks a week.
We'll pick up with a plain-English look at choosing between Claude's models next, since "which one do I even use" turns out to matter more than most people assume. If you'd like a hand figuring out which of these workflows is actually worth setting up for your business, I help with exactly that, so feel free to reach out.