Anthropic has updated its Claude Code and Claude Cowork tools with a new capability that lets the AI actually operate your Mac. It can open files, use your web browser, run developer tools, and interact with applications on your behalf. The catch: it only works on macOS, requires a paid subscription, and Anthropic is clear this is a research preview.
What Claude Mac Control Can Actually Do
When you enable computer use, Claude first checks whether it can complete your task through connected apps like Google Workspace or Slack. According to The Verge, the AI falls back to directly controlling your screen when no connector exists. That means it can open and edit files, navigate your desktop, fill out forms, and run development tools. It also works with Dispatch, a feature that lets you assign tasks from your phone and have them execute on your Mac while you are away.
The practical result: you can set up a morning briefing that runs on your desktop while you sleep, ask Claude to organize your Downloads folder, or have it run tests in your development environment and ping you when it finishes.
How the Permission System Works
Anthropic built explicit permission gates into every step. Claude asks before it opens an app, before it clicks somewhere new, and before it accesses anything outside a granted folder. You can revoke access at any time by closing the app or denying a prompt. CNET notes that the company recommends not using the feature with apps handling sensitive data like banking or medical information. Some app categories are disabled by default.
Cowork stores conversation history locally on your device rather than on Anthropic’s servers, which matters if you are thinking about what data leaves your machine.
What This Is Not
This is not a polished product. Anthropic explicitly says computer use is slower than direct integrations with apps that have APIs, and that complex tasks may need a second attempt before they work correctly. The research preview label is deliberate. Windows support is planned but not available yet.
The Bigger Picture
Anthropic is not the only company building AI agents that operate directly on your computer. Meta’s Manus launched a similar desktop app called My Computer in March. Open-source projects like OpenClaw have been doing this for months on Mac minis and personal machines. What Anthropic brings is the Claude model and its stated commitment to safety and permission-based design. As we’ve covered in our Claude review, the company’s approach to AI safety and user control is a key part of how it positions these capabilities against competitors.
How to Get It
The feature is live now for Mac users with a Claude Pro subscription at $20 per month or Claude Max at $100 to $200 per month. You will need to be on macOS and have an active subscription. Enable computer use in your Claude Desktop settings under General → Desktop app. Windows support is expected but not yet available.
