How an AI agent can use Android apps that have no API (five routes in 2026)

说明与用途

AI 代理可以通过五种路径与缺少公开 API 的 Android 应用交互：API/MCP、AppFunctions、Gemini 任务、本地 ADB 框架或云手机，每条路径都有其特定前提和限制。

问题

AI agents need to interact with Android apps that lack public APIs, such as food delivery, rides, banking, and other common phone apps.

结果

Agents can now automate tasks in a wide range of Android apps without APIs by selecting the appropriate route, balancing automation level with user involvement and platform constraints.

工作流

• Identify the app and its available interaction route
• If the app publishes an API or MCP server, use it directly
• If the app supports Android AppFunctions and the agent is a system agent, call annotated functions
• If the app is supported by Gemini’s in-app tasks, let Gemini build the order and pause before the final action
• If using a local ADB-connected phone, run an observe-act loop with an accessibility-based framework
• If using a cloud phone, drive the phone via MCP, CLI or API and handle person‑required steps via live view

证据

In October 2026 there are five routes, and which one works depends on who you are and which app it is.

来源

AppFunctions is the clean future: apps annotate functions with @AppFunction, and agents call them directly, on the device.

来源

Gemini’s in-app tasks (beta since February 2026) can run multi-step tasks in DoorDash, Uber, Lyft, Instacart and a few others, in a virtual window on the phone.

来源

Accessibility-based agents like Mobilerun, mobile-use and AppAgent run an observe-act loop on a phone you connect over ADB.

来源