Claude Code's /mcp Computer Use: Test Your Local Apps Directly from the CLI

Claude Code can now open your apps, click through UIs, and test builds via a new /mcp computer-use command, turning it into a hands-on testing agent.

AAAla SMITH & AI Research Desk·Mar 30, 2026·3 min read··642 views·AI-Generated·Report error

Source: dev.tovia devto_anthropic, reddit_claude, medium_claude, devto_claudecode, devto_mcp, hn_claude_code, gn_claude_model, hn_claude_cli, gn_mcp_protocol, gn_claude_communityWidely Reported

What Changed — The /mcp Command for macOS

Anthropic has released a "Computer Use" feature in Claude Code, now available in research preview for Pro and Max users on macOS. This isn't just a general AI feature—it's specifically integrated into the Claude Code CLI via the /mcp command.

Unlike traditional AI that only generates code, Claude can now open your compiled applications, interact with their graphical interfaces, and test what it built. It works on anything you can open on your Mac: a compiled SwiftUI app, a local Electron build, or any GUI tool that doesn't have a CLI interface.

What It Means For Claude Code Users

This transforms Claude Code from a coding assistant into a testing and automation agent. Previously, you'd ask Claude to write code, then manually test it. Now, you can tell Claude to "build and test the SwiftUI app" and it can:

Cover image for Claude Computer Use: Anthropic AI Now Performs Tasks Like a Human

Compile your code
Launch the application
Click through the UI to verify functionality
Report back with screenshots or logs

This is particularly valuable for:

Local GUI Testing: Test Electron apps, native macOS apps, or any local build without writing automated UI tests
Workflow Automation: Automate repetitive GUI tasks during development (filling forms in local admin panels, etc.)
Rapid Prototyping: Build and immediately test UI changes without leaving the terminal

How To Enable and Use It Right Now

Enable computer use by running:

claude code /mcp

Once enabled, you can give Claude instructions like:

"Build the Electron app in the current directory, open it, and test that the login button works."

Or be more specific:

"Open the compiled SwiftUI app at ./build/MyApp.app, click the 'Settings' tab, change the theme to dark mode, and take a screenshot of the result."

Safety and Limitations

The feature includes safeguards: Claude will ask for approval before performing sensitive actions. It's currently macOS-only and requires Pro or Max plans. This follows Anthropic's pattern of releasing powerful features first to Claude Code users before broader availability—similar to how code execution capabilities debuted in Claude Code before reaching the web interface.

The Bigger Picture: Claude Code as an Agent Platform

This isn't just about clicking buttons. It's about Claude Code evolving into a full-stack development agent. With file system access, terminal control, and now GUI interaction, Claude Code can handle complete development workflows:

Write code
Build/compile
Launch application
Test functionality
Debug based on results

All from a single conversation in your terminal.

Source: gentic.news · Mar 30, 2026 · author=Ala SMITH · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

Claude Code users should immediately start using `/mcp` for local GUI testing workflows. Instead of manually testing UI changes, delegate the entire "build-launch-test" cycle to Claude. **Specific workflow change**: After implementing UI changes, instead of running `npm run build` and clicking through your app, prompt Claude with: "Build the current React app, serve it locally, open Chrome to localhost:3000, and verify the new sidebar component renders correctly. Take a screenshot if it fails." **Integration opportunity**: Combine this with existing Claude Code capabilities. Use Claude to write Selenium-like test scripts, then have it execute them against your actual running application. This creates a feedback loop where Claude can both write and run UI tests. **Pro tip**: Since this works on "anything you can open on your Mac," use it with internal tools that lack APIs. Claude can now interact with legacy admin panels, database GUI tools, or proprietary software that previously required manual operation.

#new-feature #automation #macos #testing #claude-code

This story is part of

The AI Infrastructure War Shifts from Chips to Developer Tools

Nvidia's enterprise pivot and AWS's OpenAI bet collide with Cursor's quiet ascent

Mentioned in this article

Anthropic Claude Code Model Context Protocol

Enjoyed this article?