mirroir-mcp
mirroir-mcp is an MCP server that lets an AI agent observe and control a real iPhone (via macOS iPhone Mirroring). It provides tools such as describing the current screen (with OCR/icon/AI-vision backends) and executing actions like tap, swipe, and type, enabling closed-loop “observe, reason, act” workflows and skill generation/testing for mobile UI automation.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Operates locally (stdIo MCP) and relies on macOS user-granted permissions (Screen Recording/Accessibility). The README mentions API keys for optional AI diagnosis/vision and environment-variable usage for keys, but does not describe how secrets are stored/cleared or how tool execution is constrained. No information is provided about transport security, RBAC/scope enforcement, or rate limiting.
⚡ Reliability
Best When
You have a macOS 15+ machine with iPhone Mirroring enabled and want an MCP-based agent to interact with a real iPhone UI using screen understanding plus action tools.
Avoid When
You cannot grant Screen Recording and Accessibility permissions, or you need strict rate limiting/role-based access control for tool execution.
Use Cases
- • Mobile UI exploration and workflow generation for apps on a real device
- • AI-assisted end-to-end testing of iPhone screens (including deterministic skill replay)
- • Agent-driven accessibility-like interactions: tap/type based on on-screen labels and structure
- • CI/mobile testing with compiled skills (coordinate/timing capture to reduce OCR overhead)
- • Interactive debugging/diagnosis when test steps fail (optional AI diagnosis agents)
Not For
- • No-network, server-side automation at scale without a macOS host and iPhone Mirroring
- • Situations requiring strong audit/compliance guarantees for device interaction without additional controls
- • Headless environments where macOS screen recording/accessibility permissions cannot be granted
- • Use cases that need guaranteed safety boundaries (the agent can drive taps/types on a real device)
Interface
Authentication
The documentation describes local operation via stdio for MCP clients. For AI vision mode, it routes vision requests through already-authenticated CLI tools; it does not describe separate OAuth scopes for the MCP server itself.
Pricing
Pricing is not described in the provided README; costs may depend on optional AI vision/diagnosis backends and any model subscriptions/API keys used by embedded/selected agents.
Agent Metadata
Known Gotchas
- ⚠ Requires macOS Screen Recording and Accessibility permissions; first-run prompts can block tool calls until granted.
- ⚠ Vision/semantic modes depend on availability of local models (YOLO .mlmodelc) or embedded embacle FFI linkage; behavior can change based on configuration and installed components.
- ⚠ Real-device interactions are sensitive to timing and transient dialogs; generated skills may require recalibration/adjustment.
- ⚠ Exploration is bounded (max_depth/max_screens/max_time), so complete traversal is not guaranteed.
Alternatives
Full Evaluation Report
Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for mirroir-mcp.
AI-powered analysis · PDF + markdown · Delivered within 30 minutes
Package Brief
Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.
Delivered within 10 minutes
Score Monitoring
Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.
Continuous monitoring
Scores are editorial opinions as of 2026-03-30.