Claude Computer Use

See how owners enable Claude computer use, what members approve, and which desktop tasks make a good first pilot for a Team or Enterprise organization.

Published

Claude Computer Use

A warehouse lead can ask Claude to compare a stock list with a count sheet and bring back the rows that need a recount. The lead still decides which count to trust. The work they could hand over is finding the discrepancies across desktop apps their team already uses. Anthropic began rolling out computer use in beta to Team and Enterprise organizations on November 20, 2026. November release notes

What Claude can do on your screen

Anthropic describes computer use as letting Claude “see a user's screen and work in the desktop apps they allow.” Think of a colleague at your keyboard. Claude takes screenshots to see where it is, then clicks or types where the app permits it. That makes a desktop stock system with no convenient export a good candidate. If a direct connection already brings the same records into your reporting tool, keep using it. Reading a screen is slower and easier to get wrong.

The Team and Enterprise beta runs in the new Claude experience on Claude Desktop for Windows or macOS 14 and later. You can also start from Claude on the web or Claude Mobile when Computer access is on and your chat is connected to a computer with Claude Desktop open. The desktop does the work. Tasks started through the separate Cowork option and sessions in Claude Code fall outside this rollout.

Three connected clay switches show the owner enabling access, the member enabling it locally, and the member approving an app.
The owner decides what is available. The person at the computer still approves each app Claude uses.

How an owner turns it on

Computer use starts off. On Team, switching it on makes it available to every member. Enterprise role controls let you narrow the rollout before anyone gets access. Organizations with a HIPAA-ready configuration, Zero Data Retention for Claude Code, customer-managed encryption keys, or FERPA or K-12 settings are ineligible. Anthropic’s organization setup guide

  1. Sign in as an Owner or Primary Owner. Open Organization settings > Capabilities and turn on Chat and Cowork unified. Members who will use the feature need access to that experience.
  2. On Enterprise, check roles before enabling it. Built-in roles get access immediately when you switch it on. To exclude someone with a built-in role, assign them a restricted custom role and wait 15 minutes first. To exclude a custom role set to All capabilities, change it to Only selected, turn off Computer use for the role and wait 15 minutes.
  3. Under Claude Desktop, find Computer use and switch it on. If an Enterprise role list opens, clear the roles that should be excluded, then click Turn on. If the setting is missing or unavailable, check eligibility and rollout status.
  4. Under App access, choose Block these apps to deny named apps, or Allow only these apps to permit a specific set. Choose Add app, select the apps or enter their IDs, then click Done. A new allow list takes effect when you add its first app.
  5. Check the policy on a device with the app installed. Wait a minute, quit and reopen Claude Desktop, then start a new chat and ask for the app by its exact name. For a denied app, Claude should explain the restriction. Allow time across the fleet. Changing an app list or turning off computer use can take about an hour, sometimes longer, to reach each desktop.

What a member sees

  1. Open the latest Claude Desktop on the computer you want Claude to use. Under Settings > Capabilities, find Enable computer use. If it’s missing there, look under Settings > System > This computer. Turn it on for this computer and account. Hover over a grayed-out switch to see what is blocking access.
  2. On a Mac, open System Settings > Privacy & Security and grant Claude Desktop Accessibility and Screen & System Audio Recording. On macOS 27, Accessibility appears as Device Control and Data Access. Windows requires no separate operating-system permission.
  3. Start a chat in the new Claude experience. To use the web or mobile, leave the computer awake with Claude Desktop open. Your owner must also enable Computer access. Pick the computer when prompted, or use + > Devices on the web.
  4. Name the app, the source material and the result you want. Confirm the first use of computer use and review the prompt for each app. Once you approve an app, Claude can keep using it in that session. You can stop it at any time. Member setup and permission prompts

What does an app approval allow? In most approved apps, Claude has full access. Through computer use, browsers and many investment or trading apps are view-only. In terminals and code editors it can click and scroll, but typing is disabled. Some media apps are off limits. Claude is trained to check with you about actions your request leaves unclear, such as sending a message or submitting a form. Approval does not guarantee a fresh prompt before every consequential click.

Five jobs to hand over, in order

Inventory reconciliation. A warehouse supervisor could open the stock list and Friday’s count sheet, then ask, “Match by item code. Put each mismatch in a new worksheet with the code and both quantities. Leave the originals alone.” A short recount list gives the floor team somewhere to start. The supervisor can trace any row back to both sources before adjusting inventory.

A count sheet and a stock list feed a comparison station that sends a mismatched pair to a review tray.
Keep both source values with each discrepancy so the person doing the recount can check it.

Weekly operations reporting. A service manager could ask Claude to read the totals in an approved desktop scheduling app and fill a copy of the weekly staffing report. Ask it to include the date range beside each total. The manager can spend the meeting explaining a backlog and spot a previous week’s figures before sharing the report.

Catalog maintenance. A merchandising coordinator could give Claude a supplier’s public product sheet and a draft catalog workbook. Ask it to match product codes, copy the pack size into blank cells and flag unfamiliar units. A reviewer can resolve “case” versus “carton” without retyping the straightforward matches.

Presentation updates. An operations analyst could ask Claude to carry approved figures from a workbook into a copy of last week’s slide deck and list the slide titles it changed. The analyst can check the figures in context and rewrite the commentary before presenting.

Training-guide checks. A support lead could open a desktop app with sample records and ask Claude to follow a draft onboarding guide. Have it record where a button label differs and stop before saving a record. The trainer gets a specific correction to inspect. This ranks last because an unfamiliar screen may need the trainer’s judgment at each step.

What’s worth building around it

Start with a stock-discrepancy workbook. “Data entry automation software” is a commercial search term. For a distributor considering such software, the sellable piece could be a workbook that fits its stock records, paired with a reusable instruction for the team’s Claude chat. Start with item code, pack size, both source counts and a field for the supervisor’s decision. A supplier that reuses a code for different pack sizes can defeat a simple match. Flag ambiguous rows. This is the stronger first product because a buyer can check it against a completed recount.

A weekly reporting pack could serve a service manager. People also search for an “automated reporting tool.” A builder could sell a report template with agreed metric definitions, source views and a prompt that fills in a draft. The manager could compare the result with the report they signed off last week. The hard part is agreeing whether “open jobs” includes work waiting for parts. Automating an ambiguous measure just delivers the argument sooner.

Where to draw the line

An approved app can open, change or delete files outside connected folders, even folders marked read-only in Claude’s folder settings. There is no isolated workspace between Claude and the app. Screenshots are saved with the chat and can include confidential material. An automated action check can see the latest screenshot, and Anthropic may keep a copy with its record of the check. For a pilot, use sample stock records on a desktop suitable for sharing.

App lists govern computer use. They do not govern websites and extensions inside an approved browser, apps inside an approved remote desktop, or separate tools such as Claude in Chrome. If your team needs to keep computer use off selected devices, IT can enforce that through device management.

How the API differs. The Team and Enterprise feature works in the member’s Claude chat against their approved desktop apps.
With the computer-use tool in Claude’s API, a developer supplies the computer environment and their application executes Claude’s requests for screenshots, clicks and keystrokes.

For a custom implementation, the computer-use agent guide covers where screen control fits, and the multi-action cost breakdown covers developer workloads.

How do you allow Claude to use your computer?

An owner enables the unified experience and computer use for your organization and role. You enable it locally in Claude Desktop and approve each app it asks to use.

Can Claude see your screen?

Yes. It takes screenshots to navigate, and they are saved with the chat. Close confidential material before starting a task.

Can Claude on the web access your computer?

Yes, when your organization enables Computer access as well as computer use. Connect the chat to the target computer and keep Claude Desktop open there.

For a Mac-based team comparing options, see Gemini computer use on Mac and the Mac assistants guide for business workflows.

If you want a reconciliation assistant built around your team’s records, see AI agent development.

Published
Category
AI
Related Articles
Best AI Personal Assistant in 2026: Fyxer, Reclaim, Lindy and 5 More Compared

Best AI Personal Assistant in 2026: Fyxer, Reclaim, Lindy and 5 More Compared

Compare eight AI personal assistants for email, scheduling and meeting follow-ups. See what each costs, what it handles and where you still step in.Nov 21, 2026AI
Claude in Chrome: What It Is, What It Costs, and How to Start

Claude in Chrome: What It Is, What It Costs, and How to Start

Use Anthropic’s extension in the Chrome side panel, Claude Cowork, or Claude Code. See who gets access and when a browser task needs you.Nov 21, 2026AI
ChatGPT Wallet: Who Can Use It and How to Approve a Purchase

ChatGPT Wallet: Who Can Use It and How to Approve a Purchase

See who can use ChatGPT Wallet, how to add a card, what you approve before your dot pays, and how a store can recognize an agent purchase.Nov 20, 2026AI
Claude Meetings

Claude Meetings

Add Claude to Google Meet on Team or Enterprise. See what guests can see, who gets the recap, and how to use the notes after a client call.Nov 20, 2026AI
Gemini 4 Neon vs GPT-6.2

Gemini 4 Neon vs GPT-6.2

Choose between Neon, Sol and Astra using verified API rates, matched coding scores and a ticket your developers can grade blind.Nov 20, 2026AI
Gemini 4 Neon vs Carbon

Gemini 4 Neon vs Carbon

Pick a Gemini default for coding agents and knowledge work. Compare access, benchmarks and the December price change, then try a matched task on each model.Nov 20, 2026AI
ChatGPT Dots: What a Dot Is, Who Gets One and How to Start

ChatGPT Dots: What a Dot Is, Who Gets One and How to Start

Find out who can create a ChatGPT dot, where it can work between conversations, and how to set approvals, pause a project or delete its memories.Nov 20, 2026AI
GPT-6.2 vs Claude Fable 5.2

GPT-6.2 vs Claude Fable 5.2

Choose Astra, Fable or Sol for coding agents, long documents and daily work. See where their API rates diverge and which model your team can access.Nov 20, 2026AI
Newsletter

One letter, every Sunday.Working systems, not hot takes.

Weekly. No spam. Unsubscribe anytime.