Claude Desktop Officially Supports Computer Control Features: Mouse, Keyboard, Screen Fully Managed, macOS Launches First

Anthropic engineer Felix Rieseberg announced on March 23 that Claude Desktop officially launched the Computer Use feature, allowing AI to directly control the user’s mouse, keyboard, and screen, enabling it to operate any application locally. Currently, this feature is available first on macOS, with the Windows version expected to follow within a few weeks.

From API experiments to consumer features

Anthropic’s Computer Use technology previously existed in API form, mainly targeting developers. Integrating it into Claude Desktop means that ordinary users only need to download the application to let Claude take over computer operations without any programming background.

Rieseberg emphasized that the speed at which Claude performs actions is deliberately set to be slower than humans to ensure safety and controllability. Support covers both Claude Cowork and Claude Code Desktop versions.

Dispatch feature: let AI handle your computer remotely

Along with the computer control function, there is also a remote delegation mechanism called Dispatch. Users can assign tasks to Claude remotely via their phone when leaving their computer, allowing AI to continue working locally.

Rieseberg gave an example: users can instruct Claude to “automate all button sequences in an old application,” or have Claude assist with interface operations when debugging native apps.

Facing direct competition with OpenAI Operator

This release marks Anthropic’s official entry into the consumer market for “AI agent operations,” directly competing with OpenAI’s Operator feature. Both aim to enable AI to autonomously complete complex tasks on a computer, but their implementation paths differ—Anthropic integrates into desktop applications, while OpenAI primarily uses web browser control.

Security concerns and community reactions

After the feature was released, discussions about security risks intensified. Some developers pointed out that Claude can see all content on the user’s screen, making transparency of permission models and isolation mechanisms a key issue. Some users suggested running it inside a virtual machine to reduce potential risks.

Additionally, recent issues with declining quality of Claude Code services have also caused user dissatisfaction, with some developers believing Anthropic should prioritize fixing existing features rather than continuously launching new ones. The introduction of the computer control feature in Claude Desktop represents an important step from AI assistants being “dialogue tools” to “work agents.” With the macOS version launched and the Windows version soon to follow, Anthropic and OpenAI’s competition in the AI agent market has officially entered a heated stage.

This article originally appeared on Chain News ABMedia: Claude Desktop officially supports computer control—full mouse, keyboard, and screen takeover, first on macOS.