Claude Desktop can now set up its own server, open browsers, and run tests.

What happened

Developer Boris Cherny said on social media that Claude Desktop can now automatically start a local development server—using the built-in browser to test the API, view logs, and even find issues and fix bugs—without needing someone to watch over it.

What it can do

In short, it lets Claude run the entire workflow on its own: “write code → start the service → test → find issues → fix.”

• Automatically runs the full process: no need for you to manually start the service or switch to the browser to verify—Claude handles it all
• Works with existing features:
  • Visual diff
  • Permission mode (can be set to automatically accept changes)
  • Open multiple Git worktrees at the same time
  • Shared configuration, plus the recent computer use preview
• Positioning: basically for people who are used to graphical interfaces, replacing the command-line or VSCode-style workflow

Why it’s worth paying attention

• Less hassle: you tell it what you need to do, Claude writes code, tests it, finds problems, and fixes them. Previously, every step required your confirmation—now it can be automated
• AI agents go another step further: this is a more autonomous step for AI within the development workflow. It’s attractive to teams that want “stable, well-behaved, and low-maintenance” tooling, but it also means more dependence on Anthropic’s environment
• Real-world adoption: people on GitHub are interested, but how many actually use it day-to-day is not publicly known

Impact assessment

Conclusion: The trial is still at an early stage—it’s best suited for engineering teams that prioritize delivery efficiency and developers building tools; it’s not particularly relevant to people who do only trading.