Claude Code can now take over your computer
Anthropic extends Claude Code with Desktop Automation: controls mouse, keyboard and screen – Research Preview for macOS.
Anthropic has expanded Claude Code and Cowork with a Desktop Automation feature that allows the AI assistant to control your computer directly. The feature is currently available as a Research Preview for Claude Pro and Max subscribers – and for now only on macOS.
What the new Computer-Use feature can do
Claude can now use mouse, keyboard and screen to accomplish tasks that don’t have direct connectors or tools. This means:
- Open files – such as an Excel spreadsheet to extract data
- Control browsers – visit web pages, fill out forms, scrape results
- Execute dev tools – start terminal commands, run scripts
- Use apps – as long as they’re not on the “Off Limits” list
The automation runs without additional setup: Claude scrolls, clicks and navigates like a human user – but only after explicit permission.
Where the function shines: Dispatch + Cowork
The new Computer-Use feature combines seamlessly with Dispatch, a feature that lets you assign Claude tasks from anywhere – even from your smartphone. So you can give a task like “Export your last Slack messages as CSV” while on the go, and Claude will take care of it independently as soon as it gets access to your computer.
Claude prioritizes the most precise tools: First, it tries to work through existing connectors (Slack, Google Calendar, etc.). If a connector is missing, it falls back to Desktop Automation.
The limits: What Claude (still) cannot do
Anthropic has built security barriers to minimize risks like prompt injection. By default, access to certain applications is blocked:
- Investment and trading platforms
- Cryptocurrency apps
- Sensitive data operations (moving money, modifying files, scraping face images)
However, the company warns clearly: “These training safeguards are not perfect and not absolute.” Claude might occasionally act outside these boundaries – and in doing so, sees everything that appears on the screen, including personal data, sensitive documents, and private information.
Why this matters for KI Agent developers
Claude Code’s Desktop Automation marks a clear step toward true agent autonomy. Until now, developers had to provide specific APIs or plugins for each application. Now the agent can independently operate the graphical user interface – a paradigm shift that reduces tool fragmentation.
At the same time, the introduction shows how difficult it is to balance security and usability. Anthropics transparent warnings about the limits of the safeguards are a lesson for anyone who wants to deploy agents in production environments.
When you can try the feature
The Computer-Use feature is available now as a Research Preview in the Claude Desktop app for macOS. A valid Claude Pro or Max subscription is required. Windows and Linux support is expected to follow, but no timeline has been announced yet.
Takeaways for Agent Developers
- Desktop Automation becomes a core capability – agents that can interact directly with the GUI need fewer specialized connectors.
- Security remains complex – even with automatic scans and access restrictions, risks remain – especially regarding visible screen content.
- The combination with Dispatch opens new use cases – remote task assignment becomes practical, even for users without technical backgrounds.
- Anthropic is taking a gradual approach – Research Preview, macOS only, clear warnings: a conservative rollout that considers learning curves.
The expansion shows that the boundary between “assistant” and autonomous agent continues to blur. Claude can now not only write code but also control the environment where that code runs – a milestone on the way to AI-guided workflows.