I was debugging a layout issue in a Blazor app last week. The sidebar collapsed wrong at narrow widths. The kind of bug where you resize the browser window, squint at the screen, check the CSS, resize again, check again. Fifteen minutes of clicking before I even touched the code.
Then I tried something new. I told Claude Code: “Resize the browser window until the sidebar breaks, screenshot the broken state, then check the CSS for the sidebar container.”
Claude did exactly that. It resized the window, found the breakpoint where things went wrong, took a screenshot, read the CSS, and suggested a fix. All from the terminal. I never touched the mouse.
That’s Computer Use.
What Computer Use actually is
Computer Use lets Claude Code control your screen — the real screen, not a simulated one. Claude can open apps, click buttons, type text, scroll, resize windows, and take screenshots. It’s the same Computer Use API that Anthropic launched as a research preview, now integrated directly into Claude Code.
Think of it as giving Claude a pair of eyes and hands. It can see what’s on your screen (via screenshots) and interact with it (via mouse and keyboard control). This means Claude can work with any application that has a GUI — even when there’s no CLI, no API, and no MCP server available.
The key insight: Claude tries the most precise tool first. If there’s an MCP server for the service, Claude uses that. If it’s a shell command, Claude uses Bash. If it’s browser work and you have Claude in Chrome, it uses that. Computer Use is the fallback for everything else — native apps, simulators, design tools, and anything without a programmatic interface.
Why this matters for .NET developers
As a .NET developer, you live in a world of GUIs. Visual Studio. Blazor apps in the browser. MAUI apps on the desktop. The iOS Simulator. Azure Portal. These are all tools you use daily that Claude Code previously couldn’t touch.
Computer Use changes that.
Blazor development. You’re building a Blazor Server app and want to verify that a component renders correctly. Instead of manually opening the browser, navigating to the page, and checking the layout, you tell Claude: “Build the project, open it in the browser, navigate to the dashboard page, and screenshot the layout.” Claude does it all.
MAUI and native apps. Testing a .NET MAUI app on the iOS Simulator? Claude can build it, launch the simulator, tap through screens, and report what it finds. No XCTest setup, no Appium configuration. Just a natural language prompt.
Visual Studio navigation. Need Claude to check something in the Solution Explorer, verify a NuGet package version in the package manager UI, or look at a designer view? Computer Use makes that possible.
Reproducing visual bugs. “The modal is clipping on small windows” is a pain to debug from the terminal. With Computer Use, Claude resizes the window, reproduces the bug, screenshots the evidence, reads the code, and suggests a fix — all in one conversation.
How to enable it
Computer Use is available on Pro and Max plans, requires macOS, and Claude Code v2.1.85 or later. It’s off by default.
To enable it:
- Open an interactive Claude Code session
- Run
/mcp - Find
computer-usein the server list - Select it and choose Enable
This setting persists per project — you only do it once.
The first time Claude uses your screen, macOS asks for two permissions:
- Accessibility — lets Claude click, type, and scroll
- Screen Recording — lets Claude see your screen
Grant both. You may need to restart Claude Code after granting Screen Recording.
After that, just ask Claude to do something that needs the GUI:
Build the app, launch it, and click through each tab to make sure
nothing crashes. Screenshot any error states you find.
Per-app approval — you stay in control
Enabling Computer Use doesn’t give Claude free rein over your entire machine. The first time Claude needs a specific app in a session, you see a prompt showing:
- Which apps Claude wants to control
- Any extra permissions requested (like clipboard access)
- How many other apps will be hidden while Claude works
You choose Allow for this session or Deny. Approvals last only for the current session. Next session, Claude asks again.
Apps with broad reach get an extra warning:
| Warning | Applies to |
|---|---|
| Equivalent to shell access | Terminal, VS Code, Rider, and other IDEs |
| Can read or write any file | Finder |
| Can change system settings | System Settings |
These apps aren’t blocked — the warning lets you decide whether the task warrants that level of access. For most .NET development tasks, you’ll be approving your browser and maybe the iOS Simulator.
How it works under the hood
When Claude starts controlling your screen, a few things happen automatically:
Other apps are hidden. Claude only interacts with the apps you’ve approved. Everything else is hidden temporarily and restored when Claude finishes. Your terminal window stays visible but is excluded from screenshots — so Claude never sees its own output and on-screen prompt injection can’t feed back into the model.
One session at a time. Computer Use holds a machine-wide lock. If another Claude Code session is already using your screen, new attempts fail with a clear message. This prevents conflicts between parallel sessions.
You can stop at any time. A macOS notification appears: “Claude is using your computer — press Esc to stop.” Press Esc anywhere to abort immediately. Claude releases the lock, unhides your apps, and returns control to you.
Practical examples for .NET developers
Here are prompts I’ve found useful in my own workflow:
Validate a Blazor component
Run dotnet run on the BlazorApp project, open the browser to
https://localhost:5001/dashboard, resize the window to 768px wide,
and screenshot the layout. Check if the sidebar collapses correctly.
Test a MAUI app in the simulator
Build the MAUI project for iOS, launch it in the iOS Simulator,
tap through the onboarding screens, and tell me if any screen
takes more than a second to load.
Verify a desktop app
Build the WPF project, launch it, open the settings dialog,
and verify that the theme toggle switches between light and dark.
Screenshot both states.
Debug a visual regression
The data grid in the admin panel is cutting off the last column.
Open the app, navigate to the admin panel, screenshot the broken
state, then check the XAML for the grid column definitions.
What it can’t do (yet)
Computer Use is a research preview, and there are real limitations:
- macOS only in the CLI. If you’re on Windows or Linux, Computer Use in the CLI isn’t available yet. The Desktop app supports macOS and Windows.
- Pro or Max plan required. Not available on Team or Enterprise plans.
- Interactive sessions only. You can’t use Computer Use with the
-pflag for non-interactive mode. No automated CI/CD pipelines (yet). - It’s the slowest tool. Screenshots and clicks take time. For anything that can be done via CLI or API, Claude will (and should) use those faster tools instead.
- Browsers and trading platforms are view-only. Claude can see but not interact with certain sensitive apps.
- Third-party API providers not supported. You need a claude.ai account — Bedrock, Vertex, and Foundry don’t support Computer Use.
These limitations are reasonable for a research preview. The most important one to internalize: Computer Use is for tasks that nothing else can reach.
Try it yourself
Next time you’re debugging a visual issue or testing a GUI flow, try this instead of doing it by hand:
- Run
/mcpand enablecomputer-use - Grant the macOS permissions when prompted
- Ask Claude to open your app and check the thing that’s been bugging you
Start with something low-stakes — screenshot a page, verify a layout, test a button. Once you see Claude resize a window, find a bug, and suggest a fix without you touching the mouse, you’ll start thinking about all the GUI tasks you’ve been doing manually that don’t need to be manual anymore.
The terminal was always Claude Code’s world. Now your screen is too.