# Cursor plus a cloud desktop: let the agent see what it built

*Kubi, cloud desktops for AI agents. https://trykubi.com/integrations/cursor, last changed 2026-11-23. The page as Markdown; the same facts as the page itself.*

**What it is.** Cursor's agent can write the whole feature and still has no idea what it looks like. Add Kubi to mcp.json and it gets a Linux desktop with two browsers: it opens the app, clicks through the flow, reads the screen and fixes what it actually saw. You take the same screen in the dashboard when you want to try it yourself.

Cursor is the AI editor, from Anysphere. A screen is $20 a month at the founding price, or $0.035 an hour billed by the minute.

## What Cursor gains

- **The loop closes inside the editor.** Write, open, click, screenshot, fix. No copying screenshots into the chat by hand, and no describing a bug that the agent could have looked at.
- **A reviewer's machine, not yours.** The app runs on the screen with the screen's own address, so a test sign-in, a cookie or a half-broken build never lands in your own browser profile.
- **The element tree when pixels are not enough.** read_tree gives the agent the elements on screen with ids it can act on, which is steadier than aiming at coordinates and faster than guessing from a picture.

## Set it up

1. **Add the server to mcp.json.** 

```
{
  "mcpServers": {
    "kubi": { "url": "https://mcp.trykubi.com" }
  }
}
```

2. **Sign in and pick the screen.** The first call opens our sign-in page in your browser. Choose which screens the app may use and what it may do, then go back to the app. You can take it back at any time in the dashboard.
3. **Point it at your app.** A screen reaches the internet, and you can open a port on it for your own use, so the agent can load a build running on the screen itself.

The server's address is https://mcp.trykubi.com. Each app's setup is kept current at https://docs.trykubi.com/mcp/connect.

## The first thing to try

"Open http://localhost:3000 in Chromium on my Kubi screen, walk through sign-up with a throwaway address, and screenshot every step that looks wrong."

## What to know

- Ports on a screen are for your own use, which is what the terms say: a screen is not a place to host a public service.
- Cursor runs where it always ran. The screen is a machine it reaches, not where the editor lives.
- One screen is one machine. Two agents on one screen will fight over the mouse, so give each its own.

## Questions people ask

### Why not Playwright?

Use Playwright when the job is a scripted test. A screen is for the jobs a script cannot do: a desktop app, a login with a human step, a flow nobody has written a selector for yet. The docs show driving Playwright on the screen too, which gives you both.

### Does VS Code work the same way?

Yes, through the same hosted server; the only difference is where you paste the URL. We keep a page for it.

### Can two people share one screen?

People in your organization can all open it, and one person holds control at a time. The agent waits while a person has it.

## Read next

- [MCP: Connect](https://docs.trykubi.com/mcp/connect)
- [The browser on a screen](https://docs.trykubi.com/guides/browser)
- [Ports](https://docs.trykubi.com/guides/ports)
- [Pricing](https://trykubi.com/pricing)
- [Kubi compared](https://trykubi.com/compare)
- [VS Code on a screen](https://trykubi.com/integrations/vs-code)
- [Claude Code on a screen](https://trykubi.com/integrations/claude-code)
- [your own code on a screen](https://trykubi.com/integrations/your-own-code)
