lookin-mcp-peertalk

中文

An MCP (Model Context Protocol) server for Lookin — inspect and debug iOS app UI in natural language with any LLM.

Fully compatible with the official Lookin desktop client: keep using the Lookin App for visual debugging, and pair it with an LLM + MCP to query, inspect and reason about your iOS UI in plain English.

Talks to iOS devices directly over the Peertalk protocol — no HTTP proxy in the middle. Works with USB-connected real devices and the iOS Simulator.

Prerequisites: Install LookinServer

Your iOS app must integrate the LookinServer SDK. Pick either of the following:

CocoaPods

Swift project:

pod 'LookinServer', :subspecs => ['Swift'], :configurations => ['Debug']

Objective-C project:

pod 'LookinServer', :configurations => ['Debug']

Swift Package Manager

Add the following repository:

https://github.com/QMUI/LookinServer/

Note: only integrate it in the Debug configuration to avoid bloating your Release binary.

Install the MCP server

Option 1: npx (recommended)

Claude Code:

claude mcp add --scope user lookin-peertalk -- npx -y lookin-mcp-peertalk

Claude Desktop / Cursor and other MCP clients:

Add the following entry to your MCP config file:

{
  "mcpServers": {
    "lookin-peertalk": {
      "command": "npx",
      "args": ["-y", "lookin-mcp-peertalk"]
    }
  }
}

Option 2: Build from source

git clone https://github.com/nicklxz/lookin-mcp-peertalk.git
cd lookin-mcp-peertalk
npm install
npm run build

Then register it in your MCP config file:

{
  "mcpServers": {
    "lookin-peertalk": {
      "command": "node",
      "args": ["/absolute/path/to/lookin-mcp-peertalk/dist/index.js"]
    }
  }
}

Option 3: Development mode (no build step)

{
  "mcpServers": {
    "lookin-peertalk": {
      "command": "npx",
      "args": ["tsx", "/absolute/path/to/lookin-mcp-peertalk/src/index.ts"]
    }
  }
}

Available tools

Once installed, the LLM can use the following 6 tools to inspect your iOS app:

lookin_list_devices

List all connectable iOS devices (USB real devices and simulators).

• Arguments: none
• Returns: a list of devices with their type and identifier

lookin_list_apps

List all running apps that have LookinServer integrated.

• Arguments: none
• Returns: a list of apps including portKey, appName, bundleId, etc.

lookin_connect_app

Connect to a specific iOS app.

• Arguments:
  • portKey (string, required) — the app identifier returned by lookin_list_apps
• Returns: the connection result, including basic app info

lookin_get_hierarchy

Fetch the view hierarchy tree of the currently connected app.

• Arguments:
  • maxDepth (number, optional) — limit the depth of the returned tree; omit to get the full hierarchy
• Returns: a JSON view tree where each node contains oid, className, frame ([x, y, w, h]), children, etc.

lookin_get_attributes

Fetch detailed attributes for a specific view.

• Arguments:
  • oid (number, required) — the object ID of the view, taken from lookin_get_hierarchy
• Returns: a grouped list of attributes including bounds, backgroundColor, font, and many more

lookin_get_screenshot

Take a screenshot of a specific view.

• Arguments:
  • oid (number, optional) — the object ID of the view; if omitted, captures the root window
• Returns: PNG image data, base64 encoded

Usage examples

Step-by-step

1. List devices — call lookin_list_devices to see what's available
2. List apps — call lookin_list_apps to find apps running LookinServer
3. Connect — call lookin_connect_app with the portKey from step 2
4. Get the view hierarchy — call lookin_get_hierarchy
5. Inspect attributes — call lookin_get_attributes with an oid from step 4
6. Take a screenshot — call lookin_get_screenshot with an oid from step 4

Typical flow:

list_devices → list_apps → connect_app → get_hierarchy → get_attributes / get_screenshot

Natural language (recommended)

Just describe what you want in natural language and let the LLM orchestrate the tool calls:

"Connect to the iOS app that's currently running, fetch its view hierarchy, find a UILabel, then show me its attributes and a screenshot."

The LLM will chain list_devices → list_apps → connect_app → get_hierarchy → get_attributes → get_screenshot automatically — one prompt, the whole flow.

License

MIT