Claude Can Now Control Your Computer: Dispatch and Computer Use Explained

Quick Brief

• Claude now clicks, types, and navigates your screen to complete tasks inside Cowork and Claude Code, with no setup required
• Dispatch creates one continuous conversation across your phone and desktop, so you assign work remotely and return to finished results
• Computer use is in research preview for Claude Pro and Max subscribers on macOS only; Team and Enterprise plans are excluded
• Claude always requests explicit permission before accessing any new application, and you can stop it at any point

Anthropic crossed a line most AI companies have only approached: Claude can now operate your Mac independently. Available now in research preview for Pro and Max subscribers, the computer use feature inside Claude Cowork and Claude Code lets the AI click, scroll, type, and navigate your screen to complete tasks you assign, including from your phone. No setup scripts. No automation frameworks. Just a task and a result.

How Claude Uses Your Computer

Claude does not jump straight to screen control. It follows a strict three-tier tool hierarchy designed to minimize errors and maximize speed:

1. Connectors first – If a connector is available, like Gmail, Google Drive, or Slack, Claude uses it. This is the fastest and most reliable path.
2. Browser second – When no connector exists, Claude navigates Chrome to work on the task.
3. Screen interaction third – Claude uses direct mouse, keyboard, and app control only when neither option works.

This tiered approach matters in practice. Pulling messages through your Slack connector takes seconds. Navigating Slack through screen interaction takes measurably longer and carries a higher error rate. Describing tasks in terms of outcomes, rather than step-by-step instructions, gives Claude the room to pick the fastest execution path automatically.

Every action requires your permission. Before Claude accesses any new application, it presents an explicit approval prompt. You retain the ability to stop Claude mid-task at any point during execution.

Also Read Claude Just Turned Your Phone Into an AI Command Center While Millions Walked Away From ChatGPT

What Dispatch Changes

Dispatch extends computer use from a desktop-only tool into a remote delegation system. Released the week prior to March 22, 2026, Dispatch creates one persistent conversation thread that runs across your phone and your Mac simultaneously. You assign a task from your phone, redirect your attention, and return to completed work on your desktop.

Anthropic’s official use case examples are direct and practical:

• Create a morning briefing while commuting on a train
• Make changes in your IDE, run tests, and put up a pull request
• Keep a 3D printing project moving according to your initial plan
• Check emails automatically every morning or pull metrics every week

You can also spin up a Claude Cowork or Claude Code session for a report entirely from your phone. The one firm requirement: your desktop must remain awake and the Claude Desktop app must stay open. If the machine sleeps, Claude stops.

Getting Started: Exact Setup Steps

Anthropic’s official setup process for computer use is straightforward:

1. Download or update to the latest version of Claude Desktop from claude.com/download
2. Open the desktop app and go to Settings > General under the Desktop app section
3. Find the Computer use toggle and turn it on
4. Open Cowork or Claude Code in the desktop app and start a session
5. Assign Claude a task involving an app on your computer; it will request permission before accessing any app

Pairing with the mobile app for Dispatch lets you then hand off tasks from your phone after completing the above steps.

What You Can Actually Do With It

Anthropic’s verified examples of computer use tasks include:

• Pull together a competitive analysis using local files and connected tools, then compile it into a formatted report
• Open your phone simulator, interact with the app you developed, and surface UX issues
• Fill in a spreadsheet with data from multiple sources, format it, and save it to a shared folder
• Navigate internal dashboards or specialized tools that have no available connector

For developers, Dispatch paired with computer use means a build, test, and pull request workflow can run while you are away from your desk. For researchers, background data collection and report compilation runs on your machine without requiring you to stay present.

Safety: What Claude Avoids and Where Risks Remain

Computer use operates outside Cowork’s normal virtual machine sandbox. Claude interacts with your actual desktop and apps directly, not an isolated environment. Anthropic has built three specific safeguards into the system:

• Per-app permissions – Claude asks before accessing each application; investment, trading, and cryptocurrency apps are blocked by default
• App blocklist – You can add any application to a block list; any Claude request to use a blocked app is automatically denied
• Action review – The system scans for prompt injection attempts and Claude flags suspicious activity, though Anthropic notes these safeguards are not absolute

Claude takes screenshots of your screen to understand how to navigate. This means it can see any information visible, including personal data and sensitive documents belonging to you or others. Anthropic explicitly advises against using computer use to manage financial accounts, legal documents, medical information, or any apps containing personal information of others.

Current Limitations

This is a research preview, and Anthropic is transparent about where it falls short:

• Complex multi-step workflows sometimes need a second attempt
• Screen interaction is consistently slower than using a direct connector integration
• macOS only; Windows support is listed as coming soon
• Team and Enterprise plan subscribers do not have access at this time
• Actions taken in one app can affect others; clicking a link in your email app may open Chrome even without explicit Chrome permission

These are genuine constraints, not minor caveats. For workflows involving sensitive data in any category, including healthcare, finance, or personal records, Anthropic does not recommend using computer use yet.

Also Read GPT-5.4 Mini and Nano: OpenAI’s Smallest Models Just Made Big AI Affordable

Frequently Asked Questions (FAQs)