Cursor Cloud Agents Can Now Run on Windows and Linux via Cua
- Source
- x.com
- Date
1/ You can now run @cursor_ai Cloud Agent workers on Windows and Linux with Cua Cloud Fleets. Cursor runs the agent loop while your worker executes tools in a Cua VM. Launch a Cloud Fleet: https://t.co/OME0RPUXOL Here's how it works.
2/ The guide walks through a Linux worker with repository checkout and terminal tools, plus an XFCE/X11 desktop and browser. For graphical tasks, enable Cursor's computer-use option and complete its Linux setup checks.
3/ A Python controller creates a Cua pool, claims a VM, and starts the Cursor worker inside it. The worker connects outbound to Cursor and sends back tool results. Pick its Team Pool when launching a Cloud Agent run.
4/ Keep the controller running while the worker is active. Stopping it releases the VM claim; the pool keeps warm capacity for your next run. The guide covers one worker. Scaling from Cursor's queue needs an additional controller.
Context
Cursor's Cloud normally runs its worker in Cursor's own environment. Cua's guide instead has that worker run inside a virtual machine that Cua provisions, which matters for tasks that need OS-specific tools or an actual graphical desktop rather than a headless container. For Linux, the setup includes repository checkout and terminal tools plus an XFCE desktop and browser, with Cursor's computer-use option enabled for graphical tasks.
A Python controller creates a pool of Cua virtual machines, claims one, and starts the Cursor worker inside it; the worker then connects outbound to Cursor and reports tool results back, using the pool designated for Cursor's Team Pool setting. Keeping the controller running holds the machine claim, and stopping it releases that machine back to the pool's warm capacity for a later run.
The published setup covers a single worker, so matching the throughput of Cursor's own task queue requires running more than one controller in parallel. Using it requires a Cursor Enterprise account with Self-Hosted Agents enabled, a service-account API key, and Cua Fleet credentials.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.
Checking sign-in…
Loading comments…