Claude Code on a Kubi screen

Claude Code already has a shell. What it does not have is a screen: a browser it can look at, a desktop app it can click, and a machine that keeps running once you close your laptop. One MCP command gives it all three on a Kubi screen, and you can watch it work and take the mouse back whenever you want.

Through the hosted MCP server$20 a screen a month, or $0.035 an hourChecked 2026-11-23

What Claude Code gains

Claude Code is a coding agent in your terminal. These three are what change when it has a desktop of its own.

It can check its own work in a real browser

Claude Code writes the page and then looks at it: a screenshot of Chromium on the screen, not a guess from the HTML. When something is wrong it clicks through the flow the way a person would and fixes what it saw.

The work carries on with your laptop shut

A persistent screen has no idle timeout, so a long job keeps going on the machine rather than in your terminal. Come back, take the live view and read where it got to.

Logins stay where it left them

The browser profile lives on the screen's own disk, so a site you signed into once is still signed in next week. No repeating a two-factor dance at the start of every run.

Set it up

About two minutes, and nothing to install: our MCP server is hosted.

  1. Step 1

    Add the server

    One command, from any folder. It is stored for you, not for the project.

    claude mcp add --transport http kubi https://mcp.trykubi.com
  2. Step 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.

    claude mcp login kubi
  3. Step 3

    Check what it can see

    Run /mcp inside Claude Code. The Kubi server lists its tools: screenshot and act for the screen, read_tree for the elements on it, exec for commands, read_file and write_file for the disk.

The server's address is https://mcp.trykubi.com. Every app's setup is also in the docs, which is where it is kept current.

The first thing to try

Ask it this

Open Chromium on my screen, go to the page you just deployed, and tell me what the form does when I submit it empty. Use a screenshot, not the source.

What to know

  • Claude Code runs on your machine; only the screen is ours. Your model, your key, your prompt, your limits.
  • The MCP server has no tool that creates, deletes, starts or stops a screen. Those stay with people, in the dashboard or the API.
  • While you hold Take control, the agent's calls are refused until you press Release or five minutes pass with no input from you.

Questions people ask

Does this need a computer-use model?

No. The screen comes back as an ordinary image with an id, and the actions are plain tool calls, so any model Claude Code runs that can read an image can drive the screen. A model trained for computer use aims better, which is why we measure them.

Can Claude Code run on the screen itself instead?

Yes, and for a job that should outlive your laptop that is the better shape: install it on the screen and run it as a service, with its key kept as a secret for that screen. The docs have the steps.

What does it cost to leave one screen on for Claude Code?

$20 a month at the founding price for a persistent screen, the same whether it runs or is stopped. A disposable screen for short jobs is $0.035 an hour, billed by the minute.

Give your agent a desktop that stays on.

Private beta. Founding price $20 a screen a month, kept while you stay.

Get early access