Safari Web Inspector Bridge

Safari Web Inspector Bridge

An MCP server that gives AI agents the ability to inspect, observe, and automate WKWebViews on connected iOS devices via Safari Web Inspector.

Category
访问服务器

README

Safari Web Inspector Bridge

Safari Web Inspector Bridge is an MCP server that gives AI agents the same capabilities a developer gets from Safari Web Inspector — inspect, observe, and automate WKWebViews running on connected iOS devices.

Architecture

graph TD
    Agent["AI Agent<br/>Claude, Codex, etc."]
    Bridge["MCP Server:<br /><code>safari-web-inspector-bridge</code>"]
    Proxy["managed child process: <code>ios-webkit-debug-proxy</code>"]
    Device["iOS Device<br/>WKWebView"]

    Agent <-->|"MCP <code>stdio</code>"| Bridge
    Bridge <-->|"WebSocket<br/>WebKit Inspector Protocol"| Proxy
    Proxy <-->|"usbmuxd<br/>(USB / Wi-Fi)"| Device

    style Agent fill:#f0f0f0
    style Bridge fill:#f0f0f0
    style Proxy fill:#f0f0f0
    style Device fill:#f0f0f0

The server spawns ios-webkit-debug-proxy as a child process, connects to the WebKit Inspector Protocol over WebSocket, and exposes everything as MCP tools. The proxy is managed for its full lifecycle -- started on init, health-checked, auto-restarted on crash, and killed on shutdown.

Proxy Lifecycle

stateDiagram-v2
    [*] --> Checking: Server starts
    Checking --> Spawning: Binary found
    Checking --> Error: Binary missing
    Spawning --> HealthCheck: Process started
    HealthCheck --> Ready: localhost:9221 responds
    HealthCheck --> Error: Timeout (10s)
    Ready --> Crashed: Process exits
    Crashed --> Spawning: Restart (once)
    Crashed --> Error: Already retried
    Ready --> Stopped: Graceful shutdown
    Stopped --> [*]

13 MCP Tools

Discovery: list_devices | list_inspectable_pages | connect

Observation: get_url | get_dom | get_network_log | get_console_log | screenshot

Automation: navigate | execute_javascript | click_element | type_text | wait_for

Prerequisites

  • macOS required for usbmuxd and iOS device connectivity
  • ios-webkit-debug-proxy: brew install ios-webkit-debug-proxy
  • iOS: Settings › Safari › Advanced › Web Inspector: enabled
  • target app WKWebView must have isInspectable = true

Installation

git clone https://github.com/andesco/safari-web-inspector-bridge.git
cd safari-web-inspector-bridge
npm install
npm run build

Add to Claude Code

claude mcp add safari-web-inspector-bridge node /path/to/safari-web-inspector-bridge/dist/index.js

Add to any MCP client

Add to your MCP configuration file:

{
  "mcpServers": {
    "safari-web-inspector-bridge": {
      "command": "node",
      "args": ["/path/to/safari-web-inspector-bridge/dist/index.js"],
      "env": {
        "SWIB_NETWORK_CAPTURE": "true",
        "SWIB_CONSOLE_CAPTURE": "true"
      }
    }
  }
}

Tools

Device & Connection

Tool Parameters Returns
list_devices (none) [{ udid, name, os_version }]
list_inspectable_pages device_udid? string -- filter to one device [{ page_id, title, url, app_bundle_id, device_udid }]
connect page_id string -- from list_inspectable_pages { connected, page_id, url, title, warnings? }

Observation

Tool Parameters Returns
get_url (none) { url }
get_dom selector? string -- CSS selector (default: document.documentElement); outer_html? boolean (default: true) -- outerHTML vs textContent { html } or { text }
get_network_log clear? boolean (default: false); filter_url? string -- regex; filter_status? string -- e.g. "302", "4xx" [{ request_id, method, url, status, mime_type, response_headers, request_headers, redirected_from, redirected_to, timing, error }]
get_console_log clear? boolean (default: false); level? "log" | "warn" | "error" | "info" [{ level, text, timestamp, source_url, line_number }]
screenshot (none) MCP image content (base64 PNG)

Automation

Tool Parameters Returns
navigate url string { url, title, status }
execute_javascript expression string; await_promise? boolean (default: true) { result } -- JSON-serialized return value
click_element selector string -- CSS selector; index? number (default: 0) -- which match to click { clicked, selector, tag_name }
type_text text string; selector? string -- focus this element first { typed: true }
wait_for One of: selector? string, url_contains? string, network_idle? number (ms); plus timeout_ms? number (default: 10000) { matched: true, elapsed_ms }

Configuration

Env Var Default Description
SWIB_AUTO_CONNECT false Auto-connect to the first inspectable page on startup
SWIB_NETWORK_CAPTURE true Capture network requests on connect
SWIB_CONSOLE_CAPTURE true Capture console messages on connect
SWIB_PROXY_PORT 9222 Starting port for ios-webkit-debug-proxy device ports

[!note] SWIB_NETWORK_CAPTURE and SWIB_CONSOLE_CAPTURE default to true — set to false to disable. SWIB_AUTO_CONNECT defaults to false — set to true to enable.

Example Workflow

sequenceDiagram
    participant Agent as AI Agent
    participant Bridge as MCP Server
    participant Proxy as ios-webkit-debug-proxy
    participant Device as iOS WKWebView

    Agent->>Bridge: list_inspectable_pages()
    Bridge->>Proxy: GET /json (devices + pages)
    Proxy-->>Bridge: page_id: "1", title: "Banks"
    Bridge-->>Agent: [{ page_id, title, url, app_bundle_id }]

    Agent->>Bridge: connect({ page_id: "1" })
    Bridge->>Proxy: WebSocket connect
    Proxy->>Device: WebKit Inspector attach
    Device-->>Proxy: Connected
    Bridge-->>Agent: { connected: true }

    Agent->>Bridge: get_network_log({ filter_url: "scotiabank" })
    Bridge-->>Agent: [{ url, status: 302, redirected_to: "scotiabank://..." }]

    Agent->>Bridge: execute_javascript("document.title")
    Bridge->>Device: Runtime.evaluate
    Device-->>Bridge: "Banks"
    Bridge-->>Agent: { result: "Banks" }

    Agent->>Bridge: click_element({ selector: "button.next" })
    Bridge->>Device: Runtime.evaluate (querySelector + click)
    Device-->>Bridge: clicked
    Bridge-->>Agent: { clicked: true, tag_name: "button" }

    Agent->>Bridge: wait_for({ url_contains: "/dashboard" })
    loop Poll until match
        Bridge->>Device: Runtime.evaluate (location.href)
        Device-->>Bridge: current URL
    end
    Bridge-->>Agent: { matched: true, elapsed_ms: 1230 }

    Agent->>Bridge: screenshot()
    Bridge->>Device: Page.snapshotRect
    Device-->>Bridge: base64 PNG
    Bridge-->>Agent: image content

Development

npm run build        # Compile TypeScript to dist/
npm run dev          # Watch mode (tsc --watch)
npm test             # Run tests (vitest)
npm run test:watch   # Watch mode tests
npm start            # Run the MCP server (node dist/index.js)

Project structure

src/
  index.ts               # Entry point, server setup, lifecycle
  types.ts               # Interfaces and config loader
  proxy-manager.ts       # Spawns and manages ios-webkit-debug-proxy
  device-discovery.ts    # Queries proxy for devices and pages
  webkit-connection.ts   # WebSocket connection to WebKit Inspector Protocol
  network-buffer.ts      # Ring buffer for network request entries (1000 max)
  tools/
    device-tools.ts      # list_devices, list_inspectable_pages, connect
    observation-tools.ts # get_url, get_dom, get_network_log, get_console_log, screenshot
    automation-tools.ts  # navigate, execute_javascript, click_element, type_text, wait_for

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选