Computer Use Feature

Feature request for product/service

Cursor IDE

Describe the request

It would be great if Cursor Agent mode have feature computer use similar as codex and claude code desktop

Operating System (if it applies)

Windows 10/11
MacOS
Linux

Hey, thanks for the feature request.

This is partially available already. In the IDE, Agent has a browser tool, so it can control the built-in browser navigation, screenshots, clicks, and self-verify tasks on the web. More details here: Browser | Cursor Docs

Full local computer use controlling your whole desktop and any apps on your machine like Codex CUA or Claude Code desktop is not available natively yet. As a workaround, you can connect a computer-use MCP server. Via MCP, Agent can get access to those capabilities: Model Context Protocol (MCP) | Cursor Docs

I’ve logged the request for a native local computer-use mode. There’s no ETA yet. If you want, you can follow a related thread where this idea is already being discussed: Feature Request: Computer Control / Coworker Mode for Cursor

Hi Dean, do you know when this feature will be available? The computer use mcp is very slow and uses a bunch of tokens aka its extremely inefficient. I notice that I really only have the chatgpt subscription right now bc that is good with computer use.

Also, codex/gpt app has a great remote control feature so I can easily see the sessions on each device and work from them. I know there is a manual remote control setup but do you know if this will be added shortly?

Both would be huge adds as they are almost necessary for agentic workflows.

Feature request for product/service

Cursor IDE

Describe the request

Cursor still has no Computer Use. Codex already does: look at the screen, click, type, and finish UI workflows in the running app. In-IDE browser (DOM only) is not the same. Without Computer Use, Cursor is far behind Codex for agent work that is not just editing files. Please ship Computer Use.

Operating System (if it applies)

Windows 10/11

Feature request for product/service

Cursor IDE / Agent

Describe the request

Please ship native local Computer Use for Cursor Agent: see the screen, click, type, and finish real UI workflows on the user’s machine — not only edit files, and not only drive the in-IDE Browser.

Related thread: Computer Use Feature

Why this matters

Coding agents are no longer just “write a patch and hope.” The high-value loop is:

  1. change code
  2. run / open the real app
  3. click through the UI
  4. verify the bug is fixed
  5. iterate

Without Computer Use, that loop breaks outside the editor. Users end up babysitting screenshots, copying errors by hand, or wiring slow third-party Computer-use MCPs. That is not competitive for an Agent-first IDE.

In-IDE Browser helps for web apps, but it is not Computer Use. Many workflows live in desktop apps, system dialogs, installers, native clients, multi-window setups, or “show me what actually happened on screen.” Those need real desktop perception + control.

Grok is especially suited for this

Grok models (and Grok-style agent workflows) are a strong fit for Computer Use:

  • long, tool-heavy, multi-step agent loops
  • noisy visual / UI feedback
  • “keep going until the task is done” behavior

Cursor already invests in strong models and Agent UX. Pairing that with native Computer Use — especially when running Grok / similar agentic models — would unlock a class of tasks that today feel stuck: UI regressions, onboarding flows, settings pages, non-web tools, and end-to-end verification.

Right now the model can reason well about the next click, but Cursor often cannot take that click on the real desktop. That gap wastes the model’s strength.

Everyone else already moved

Native or near-native Computer Use / desktop agent control has been available for a while in competing products (e.g. Codex / ChatGPT computer use, Claude Code desktop-style flows, and various agent desktops). Users notice.

Cursor’s answer today — “use Browser” or “wire an MCP” — is not the same product. MCP Computer Use is typically slower, token-heavy, and fragile. Browser is DOM-scoped. Competitors already treat desktop Computer Use as a first-class Agent capability.

If Cursor wants to stay the default Agent IDE, native Computer Use should not be a someday wishlist item. It should be core.

Ask

  1. Native local Computer Use (view + control) for Agent on desktop OS
  2. First-class support when using agentic models such as Grok — not a bolted-on MCP afterthought
  3. Clear product distinction: Browser ≠ Computer Use
  4. Public ETA or roadmap signal (even if phased: view-only → control)

Operating System

MacOS (also relevant for Windows / Linux)

Seriously, CodeX outperforms every task that I do in Cursor because of this feature. How is this not a higher priority. There is nobody taking this request seriously.