# Connect your agent to a desktop

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

Every way an agent reaches a Kubi screen. An agent app connects through the hosted MCP server at https://mcp.trykubi.com with one URL and no code; your own code uses the TypeScript SDK or the API. All of them reach the same screens, with the same permissions and the same history.

- **[Claude Code](https://trykubi.com/integrations/claude-code)** (Anthropic, a coding agent in your terminal). 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.
- **[claude.ai and Claude Desktop](https://trykubi.com/integrations/claude-desktop)** (Anthropic, the Claude app, in a browser or on your desktop). In the Claude app there is nothing outside the conversation: no shell, no browser, no files that stay. A Kubi screen is the computer it has been missing, added as a custom connector in about two minutes, with no code at all. You watch the desktop live while Claude works on it, and take the mouse the moment you want to.
- **[Codex](https://trykubi.com/integrations/codex)** (OpenAI, OpenAI's coding agent, in the terminal or the cloud). Codex is good at writing and running code and has nowhere to look at the result. Its own sandbox is made fresh and thrown away, so nothing it signed into is there next time. Two commands point it at a Kubi screen instead: a desktop with Chromium and Firefox, a disk that stays, and a person who can watch and take over.
- **[Cursor](https://trykubi.com/integrations/cursor)** (Anysphere, the AI editor). 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.
- **[VS Code](https://trykubi.com/integrations/vs-code)** (Microsoft, the editor, with its agent mode). VS Code's agent mode speaks MCP, so a desktop is one Command Palette entry away. Kubi hosts the server, so there is no process to run beside your editor and nothing to update. The agent gets a screen it can look at and act on, and you keep the live view and the mouse.
- **[your own code](https://trykubi.com/integrations/your-own-code)** (the SDK and the API, TypeScript, or any language over HTTP). When the agent is yours, the screen is an API: ask for a screenshot, send actions against the id it came with, run a command, move a file. The SDK is TypeScript; everything else talks to the same HTTP endpoints with a key in the header. One screen or a hundred, the same two calls.

## Which way in

| You have | Use |
|---|---|
| An agent app: Claude Code, claude.ai, Codex, Cursor, VS Code | The hosted MCP server |
| Your own code in TypeScript | The SDK, `@trykubi/sdk` |
| Your own code in any other language | The HTTP API |
| An agent that should outlive your laptop | Install it on the screen itself |

An agent app not listed works too if it speaks MCP: the address is the whole setup. Agents we have not run ourselves, OpenClaw and Hermes Agent among them, get a page once we have.
