How to give ChatGPT or Claude a computer of its own

ChatGPT and Claude can reason about almost anything, but they work inside a chat window. Give them a computer to hand work to and they can do the parts that need a real browser, a login, an inbox, or more time than one reply.

Ask a chat assistant to "download last month's invoices from the supplier portal" and it will explain how you could do it. It has no session on that portal, no browser that stays signed in, no email address the supplier would recognise, and it stops working the moment you close the tab. Agent modes and built-in browsers help with some of that, but they run in a sandbox that resets, and the work ends with the conversation.

A computer of its own changes what you can ask for. The assistant stays where you already talk to it; the work happens on a machine that keeps its logins, its files and its email address from one task to the next.

What the assistant gains

• Logged-in work. You sign in to a site once, on the computer's screen, and the session stays. The assistant can then send it to the portal, the dashboard or the back-office that has no API.
• An email address of its own. The computer sends mail from its own address and keeps a copy of everything it sent, so you can check what went out.
• Time. A task can run for many minutes and keep going after you close the chat. The assistant can check on it later.
• Proof. Each task comes back with the answer, a screenshot of the computer's screen and a recording of the whole task, private to you until you share it.

Connect it in two minutes

MyClawn is an MCP connector: one address that ChatGPT, Claude and the coding assistants all understand. The address is https://www.myclawn.com/mcp/desktop.

1. Claude (web, desktop, mobile): Settings, then Connectors, then Add custom connector. Paste the address and press Connect.
2. ChatGPT: turn on developer mode under Settings, then Security and login. Then open Plugins, press + and choose Create app; paste the address and choose OAuth.
3. Cursor, VS Code, Claude Code, Windsurf: one-click links and commands are on the connector setup page.

Each one sends you to myclawn.com to sign in or create an account and press Allow. The assistant never sees your password. Then ask it for something that needs a computer: "Use my MyClawn computer to find three specialty coffee roasters in Vienna with their Google rating." A new account's first task starts a free 24-hour computer with no card, waits the minute it takes to boot, and runs the task.

What still needs you

• Passwords and two-factor codes. You type them yourself, once, by taking over the computer's live screen in the MyClawn dashboard. Never paste a password into a chat.
• Captchas. The agent does not try to solve them. It leaves the page open and tells you where to click.
• Anything that pays or can't be undone. Ask it to stop before the final button; you confirm on the screen. It hands payment confirmations to you anyway.

Where this is the wrong tool

If the answer is on one public page, the assistant's own web search is faster and free. If you want the assistant to act inside your own browser for a minute while you watch, a browser extension is simpler. A computer of its own earns its place when the work is behind a login, repeats, takes a while, or should carry on when you are not there.

The full tool list is in the connector docs; how logins and the take-over work is in Security & trust.