[FEATURE] First-party desktop control (screen, mouse, keyboard) for any application

Summary

Please ship a first-party desktop-control capability for Claude Code on Windows and
macOS: screen viewing plus mouse and keyboard control of any desktop
application, with the product's permission model built in.

Why

Third-party MCP servers such as Desktop Commander give a persistent shell and file
access, but they can't see the screen or drive applications. Claude in Chrome covers
Chrome only. Much real work lives in desktop apps with no API: Plex's server UI, a
backup tool, Bitdefender settings, media and office apps. Today the agent has to hand
those steps back to the user.

What I'm asking for

• Screenshots of a chosen window or the whole screen, plus mouse and keyboard input,

working with any app.

• Per-app permission on first use, a visible indicator whenever the agent is in

control, and one key to take control back instantly.

• The same safety rules as the browser tools: no entering credentials or payments,

and confirmation before anything irreversible.

• First-party maintenance, so users aren't trusting a third-party npm package with

control of their whole desktop.

(Also: if this lands, I'll happily beta-test it. A free month of Max 20x would help
me put it through its paces. 😄)