- Demo GIF in README showing Calculator automation via Claude Code
Thanks for the feedback everyone!
ktpsns 3 days ago [-]
I wonder whether this requires particular GUI toolkits to be used, such as WFC. In any GUI there are enough "bad boy" toolkits which just "draw lines" and thus are not accessible at all.
stif1337 2 days ago [-]
No toolkit dependency, OculOS reads the OS-level accessibility tree, which is toolkit-agnostic:
The coverage varies by toolkit. Win32/WPF/GTK expose rich trees. Electron apps expose key elements but the tree is shallower. Custom-drawn UIs (games, OpenGL) have minimal or no accessibility tree. That's the main limitation.
Frannky 3 days ago [-]
Cool idea! Does this work with electron apps? I tried automating some apps and the problem was that a lot of stuff was never reachable, only via screenshot + click
stif1337 2 days ago [-]
Yes, electron apps expose a reasonable accessibility tree through Chromium's UIA/AT-SPI bridge. We've tested with Spotify (Electron/CEF), VS Code, Slack, and Chrome itself.
The tree is shallower than native Win32/WPF apps, but key interactive elements (buttons, inputs, lists) are usually exposed. You can check what's available with:
Maybe I need to familiarize myself with MCP, but wouldn't this make way more sense as a simple CLI tool instead of an HTTP service with a REST API?
stif1337 2 days ago [-]
It actually supports both you can use it as a plain REST API (no MCP needed) with any HTTP client:
curl localhost:7878/windows
curl -X POST localhost:7878/interact/{id}/click
MCP mode is an optional layer for AI agents (Claude, Cursor, etc.) that already speak MCP. The REST API works standalone for scripts, testing, CI/CD — no AI required.
lioeters 3 days ago [-]
Good idea, I'm looking forward to seeing it grow - especially the Python and TypeScript bindings.
stif1337 2 days ago [-]
Thanks! Python and TypeScript SDKs are high on the roadmap.
- Screenshot capture: GET /windows/{pid}/screenshot → returns PNG
- Batch operations: POST /interact/batch → multiple actions per request
- Wait/poll: GET /windows/{pid}/wait?q=Submit&timeout=5000
- Python & TypeScript SDKs (local install, PyPI/npm coming soon)
- OpenAPI spec, Dockerfile, 7 example scripts
- Demo GIF in README showing Calculator automation via Claude Code
Thanks for the feedback everyone!
- Windows: UI Automation (works with Win32, WPF, WinForms, Qt, Electron) - Linux: AT-SPI2 (GTK, Qt, Electron) - macOS: AXUIElement (Cocoa, Qt, Electron)
The coverage varies by toolkit. Win32/WPF/GTK expose rich trees. Electron apps expose key elements but the tree is shallower. Custom-drawn UIs (games, OpenGL) have minimal or no accessibility tree. That's the main limitation.
The tree is shallower than native Win32/WPF apps, but key interactive elements (buttons, inputs, lists) are usually exposed. You can check what's available with: