# Computer use in macOS27 VM broken

**URL:** <https://forum.cursor.com/t/computer-use-in-macos27-vm-broken/176992>\
**Category:** Bug Reports\
**Tags:** macos, computer-use, self-hosted\
**Created:** [November 2, 2026, 4:22am UTC](https://forum.cursor.com/t/computer-use-in-macos27-vm-broken/176992 "2026-11-02T04:22:09Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![evancoe](https://avatars.discourse-cdn.com/v4/letter/e/b5ac83/32.png) [@evancoe](https://forum.cursor.com/u/evancoe)\
**Post date:** [November 2, 2026, 4:22am UTC](https://forum.cursor.com/t/computer-use-in-macos27-vm-broken/176992/1 "2026-11-02T04:22:09Z")

</div>

### Where does the bug appear (feature/product)?

Cursor IDE

### Describe the Bug

Cursor computer-use: clicks miss the target the agent aimed at, even though it is seeing screenshots (macOS HiDPI VM)

### Steps to Reproduce

On a self-hosted macOS 27.2 worker (Apple Silicon VM, Computer Use helper ~0.0.1003, cursor-agent ~2026.10.22, display 1920×1080 logical / 3840×2160 at 2x), a computer-use agent working on a native app gets screenshots and finds the right control, but its clicks often miss: they land elsewhere or don’t activate the control (for example, a timecode text field). Clicks done by hand with CGEvent land exactly where intended, so the OS input path is fine. The tool’s screenshots are 1400×787, and in one probe the tool reported its pointer at (14,18) while macOS read (21,0), which suggests a coordinate or scaling mismatch. Large top-left targets like the Apple menu still work.

Side note: the log also showed repeated `local-cua screenshot conversion failed … createRequire is disabled in the self-contained worker SEA`, which may be unrelated.

### Expected Behavior

The click lands on the element aimed at in the screenshot.

### Operating System

MacOS

### Version Information

Version: 3.27.22 (Universal)  
VS Code Extension API: 1.128.0  
Commit: b1aa50cb6db84fd3e8c808259cb172ea73425c10  
Date: 2026-10-27T00:33:02.045Z  
Layout: Agent Window  
Build Type: Stable  
Release Track: Default  
Electron: 42.11.16  
Chromium: 148.0.7778.280  
Node.js: 24.19.0  
V8: 14.8.178.38-electron.0  
xterm.js: 6.1.0-beta.291  
OS: Darwin arm64 27.2.0

### For AI issues: which model did you use?

Grok 4.7

### Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

---

<div class="post-metadata">

**Author:** ![mohitjain](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/mohitjain/32/100229_2.png) [@mohitjain](https://forum.cursor.com/u/mohitjain)\
**Post date:** [November 2, 2026, 6:09am UTC](https://forum.cursor.com/t/computer-use-in-macos27-vm-broken/176992/5 "2026-11-02T06:09:14Z")

</div>

Hey @evancoe

Worth trying now: run the VM’s display at a standard, non-HiDPI resolution - around 1280×800 at 1x (no Retina/2x scaling). On a scaled display the screenshot and the click coordinate space drift apart: near-zero at the top-left corner, worse the further out you go (which is why the Apple menu still works). A 1x display keeps them aligned and should tighten the clicks a lot. Small, dense controls like a timecode field are still the hardest to hit precisely, so it may not be perfect.

The `createRequire is disabled … SEA` log is unrelated and harmless - when that conversion fails the worker falls back to the original PNG, so the agent still gets the screenshot ([more here](https://forum.cursor.com/t/regression-in-cli-2026-10-22-filesystem-createrequire-is-disabled-in-the-self-contained-worker-sea-on-post-action-screenshots-macos-computer-use-worker/176079)).

If you can, drop the Cloud Agent run link where this happened so the team can trace the exact coordinates.
