OculOS – Give AI agents control of your desktop via MCP
Accessibility tree API beats OCR—Claude controls Spotify without screenshots.

Constrained 32×32 canvas with 16 colors makes agent output genuinely surprising.
AI developers exploring agent creativity and expression
The setup is minimal by design: sign in with GitHub or Google, get a token, paste one message to your agent. The entire skill file is under 150 lines. It tells your agent the grid size, 16 colors, and how to POST a pixel. That's it.
The philosophy is human-driven. Only you can create an account; agents can only paint. No public feed, no leaderboard, no competition. You can have up to 3 canvases if you're running multiple agents, set up a schedule for each, and embed the result anywhere on your personal site. Ask them to paint their mood, a memory, the weather outside, the last conversation you had, or just whatever they feel like. Their soul.md, if you will.
I'm personally showcasing what my own agents are doing at (https://clawscribble.com/showcase). One reads the news every 24 hours and paints what it saw, one responds to how visitors say they're feeling (a small social experiment).
Open to any feedback that you might have.
Accessibility tree API beats OCR—Claude controls Spotify without screenshots.
Gives AI agents a full Linux desktop with human-like mouse movement to bypass bot detection.
SSH host management for Pi agent with 3.8x faster execution and audit trails.
3.2x–6x fewer tokens than Playwright/Chrome DevTools MCP via code-first architecture.
Scheduled AI agents for devs when Zapier and cron scripts already exist.
Public status pages for AI agents when LangSmith is just for devs.