Mirroir MCP, developed by Jfarcand, is a free MCP server tool built primarily with Swift, incorporating a Rust runtime for its AI vision capabilities. It enables AI agents to interact with and control real iPhones through macOS iPhone Mirroring, with experimental support for macOS window control. The tool captures mirrored screen content, performs OCR and BFS exploration, then injects input using macOS CGEvent API or Karabiner DriverKit virtual HID devices. It is distributed via npm, Homebrew, a shell script, and prebuilt binaries, making it accessible for developers building AI agents.

What does Mirroir MCP connect to?

Mirroir MCP functions as an MCP server tool designed to connect to and control a physical iPhone using macOS iPhone Mirroring. It also offers experimental support for managing macOS windows directly. The system operates by capturing the mirrored screen display, then applying Optical Character Recognition (OCR) and Breadth-First Search (BFS) exploration to understand the interface. Input is injected into the target device through the macOS CGEvent API or specialized Karabiner DriverKit virtual HID devices, allowing AI agents to interact with applications.

Core features and tools

Mirroir MCP offers 33 callable tools for diverse interactions. Key input tools include tap, type_text, swipe, press_key, and shake. Observation capabilities are provided by screenshot, describe_screen, start_recording, stop_recording, get_orientation, status, and classify_screen. Navigation tools such as launch_app, open_url, press_home, press_app_switcher, press_back, spotlight, scroll_to, reset_app, set_network, measure, generate_skill, list_skills, get_skill, record_step, save_compiled, and switch_target enable application control.

Who is it for?

Mirroir MCP is designed for AI agents and developers who are building these agents. It provides the necessary interface for AI agents to interact with real iOS applications and macOS windows. This allows for advanced automation and testing scenarios where direct control over mobile and desktop environments is required. Compatible AI clients include Claude Code, ChatGPT, Cursor, and GitHub Copilot (VS Code), which use the tool's capabilities. Skip this if your project does not involve AI agents controlling iOS applications or macOS windows.

How to get started

The server authenticates using a "fail-closed" model, where read-only tools are exposed by default without a configuration file. Client-server communication occurs via the standard stdio transport protocol.

Use cases

• Testing login flows: An AI agent can navigate an app, input credentials, and verify the outcome.
• Cross-app workflows: For example, getting a commute ETA from one app and then sending it via iMessage.
• Recording bug reproduction videos: An AI can perform a sequence of actions on the iPhone, capturing a video of the process.

Mirroir MCP offers a powerful solution for AI agents needing to interact with iOS and macOS environments. Its ability to capture screens, perform OCR, and inject precise inputs makes it a valuable tool for automation and testing. The "fail-closed" authentication model adds a layer of control over tool access. Requirements include macOS 15+, an iPhone with Mirroring, and specific system permissions. Try testing login flows by having an AI agent navigate an application and input credentials.