[BUG] Computer-use `zoom` returns interpolated (blurry) crops on high-res displays — small text unreadable

Environment

• Claude Code version: 2.1.198 (desktop app on Windows)
• OS: Windows 11 Home 10.0.26220
• Display: 5120×1440 ultrawide (32:9), 100% scaling
• Tools: computer-use MCP (screenshot / zoom)

Problem

The zoom tool is documented as "Take a higher-resolution screenshot of a specific region of the last full-screen screenshot. Use this liberally to inspect small text, button labels, or fine UI details that are hard to read in the downsampled full-screen image."

On a high-resolution display it does not deliver that. screenshot downsamples the 5120×1440 desktop to ~1568×441 before returning it to the model (expected — image budget). But zoom then appears to interpolate/upscale from that already-downsampled capture instead of re-capturing the requested region from the screen at native resolution. The returned crop has roughly native pixel dimensions, yet its content is soft interpolated mush: a ~3.3× downsample destroys 12px UI text, and no upscaling can bring it back.

Steps to reproduce

1. Windows machine with a large/high-res display (e.g. 5120×1440 at 100% scaling).
2. Open any window with standard 9pt UI text — e.g. File Explorer in details view.
3. Agent calls screenshot → returns a ~1568px-wide image (3.3× downsample).
4. Agent calls zoom on the file-list region.
5. Returned image: file names are unreadable smears. A true native-resolution crop of the same region would be perfectly crisp (plain Segoe UI text on screen).

Expected behavior

zoom should re-capture the requested region from the display at native resolution (or at least apply its downsampling budget per-region), since its entire purpose is recovering fine detail that the full-screen downsample lost.

Actual behavior

zoom output is visually indistinguishable from cropping + upscaling the downsampled full screenshot. Small UI text stays unreadable no matter how tight the zoom region is.

Impact

On large displays the agent effectively cannot read any small UI text: verification workflows (reading labels, dialogs, logs, file names) fail, and the agent wastes tool calls trying ever-smaller zoom regions that are all equally blurry.

---

<img width="1036" height="456" alt="Image" src="https://github.com/user-attachments/assets/58553ae1-1ebd-4f84-bae9-c10676a21f3b" />

*Drafted with Claude Code at my request, then adjusted manually