Show HN: Mirroir – MCP server that gives AI agents a real iPhone to control
It works by hijacking macOS iPhone Mirroring: the MCP server captures the mirrored screen, runs Apple Vision OCR to find UI elements with tap coordinates, then sends input through a Karabiner DriverKit virtual keyboard/mouse. The AI sees the screen, decides what to tap/type/swipe, and the iPhone responds.
Some interesting technical rabbit holes:
- iOS elements aren't exposed via Accessibility — the mirroring window is a single opaque AXHostingView with zero children. OCR is the only way to read the screen. - Clipboard paste doesn't work programmatically. At all. We tried everything — HID Cmd+V, AX menu automation, AppleScript, CGEvents. The paste bridge lives deep in the Continuity/Handoff stack and requires a physical user gesture. - Input goes through a Karabiner DriverKit extension that presents as a US ANSI keyboard to iOS, regardless of your Mac's keyboard layout.
Limitations I should be upfront about: macOS 15+ only, one phone at a time, no clipboard bridge for paste, and the phone needs to stay unlocked during a session.
It's open source (Apache 2.0): https://github.com/jfarcand/iphone-mirroir-mcp
Properly configure permissions if you use it with OpenClaw ;-)
Two reliability questions: 1) How do you handle UI drift / A-B layouts / dynamic text (e.g., different font sizes, localization)? Do you anchor on vision+OCR only, or do you also incorporate accessibility tree / UIAutomation hints where possible? 2) Do you have a concept of checkpoints + recovery (e.g., if a tap misfires, can the agent detect it’s in the wrong screen and roll back / retry)?
I’m working on SkillForge (https://skillforge.expert) — we’re doing something adjacent for web UIs: record a workflow once, then replay it later as a reusable skill with checkpoints + retries. Curious if you’ve considered exporting a “skill file” for iPhone flows (step evidence + approved boundaries), so people can share/reuse automations safely.