Invi Labs iPhone MCP

託管式 MCP 伺服器,讓你的 AI 代理能操作真正的 iPhone。它可以擷取螢幕截圖、點按、滑動、輸入文字、開啟 App,並使用自己的模型讀取螢幕內容。

託管 MCP 伺服器

npx add-mcp 'https://api.invilabs.io/mcp'

可安裝到 Claude Code、Codex、Cursor 等客戶端

文件

Invi Labs for developers and AI agents

Control real iPhones from your AI agent or your Python code. Both work with phones connected by Invi Dongle, phones linked with Invi Link, and hosted rentals.

Hosted MCP server

Address: https://api.invilabs.io/mcp. Every request sends your API key as Authorization: Bearer YOUR_API_KEY. The server has no AI of its own and never spends AI credits: your agent reads the screenshots with its own model and decides where to tap.

Claude Code

claude mcp add --transport http --scope user invilabs https://api.invilabs.io/mcp --header "Authorization: Bearer YOUR_API_KEY"

Cursor, Windsurf and Gemini CLI

Add to ~/.cursor/mcp.json (Cursor, field url), ~/.codeium/windsurf/mcp_config.json (Windsurf, field serverUrl) or ~/.gemini/settings.json (Gemini CLI, field httpUrl):

{
  "mcpServers": {
    "invilabs": {
      "url": "https://api.invilabs.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Codex

Add to ~/.codex/config.toml:

[mcp_servers.invilabs]
url = "https://api.invilabs.io/mcp"
http_headers = { Authorization = "Bearer YOUR_API_KEY" }
tool_timeout_sec = 300

Limit phones and tools

  • ?phone=SE001 — only this phone (repeat for more).
  • ?read_only=1 — only tools that change nothing.
  • ?tools=phone_*,devices — only matching tools.
  • ?exclude_tools=jb_* — hide tools.

API keys

In the dashboard (https://panel.invilabs.io), open Resources → AI agents (MCP). It creates a key with the right permission (device.control; add device.edit_proxy to let the agent change proxy and VPN settings) and gives a ready-to-paste setup for your app.

Python SDK

pip install invilabs
from invilabs import Client

client = Client("YOUR_API_KEY")
for device in client.devices():
    print(device["id"], device["name"], device["device_type"])

phone = client.phone("SE001")  # the phone's code from the dashboard
print(phone.details()["online"])
phone.type("hello from Python")
phone.press("enter")

Documentation: https://sdk.invilabs.io · MCP guide: https://sdk.invilabs.io/guides/iphone-ai-agent-mcp/ · Quickstart: https://sdk.invilabs.io/quickstart/