# mobilerun [Health: Active]

**Category:** 🖥️ OS Automation  
**Repository:** https://github.com/droidrun/mobilerun  
**GitHub Stars:** 9709  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/mobilerun

## Description
Control real Android and iOS devices with LLM agents — tap, swipe, type, automate flows.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "mobilerun": {
    "command": "uvx",
    "args": ["mobilerun"]
  }
}
```

## Documentation

## What mobilerun MCP server does

The mobilerun MCP server gives language-model agents a way to operate physical Android and iOS devices. Agents can inspect the current interface, interpret screenshots, launch apps, enter text, perform taps and swipes, and complete workflows that span several actions. Results can be returned through command-line runs or Python-based automation.

The framework is intended for local execution. It is separate from Mobilerun Cloud, which provides managed infrastructure, hosted devices, dashboards, REST APIs, SDKs, and cloud workflows. Choose the local framework when the agent and target phone should remain under your own runtime control.

## How it works

Mobilerun uses the device's Portal runtime to expose interaction and state to the agent. On Android, the Portal app works with accessibility information and screenshots to describe the current UI. Vision mode sends screenshots to the model, while vision-only mode supports applications that do not expose useful accessibility-tree data.

Tasks may run directly or with reasoning enabled. Direct runs suit simple commands, while reasoning mode uses manager-executor planning for longer sequences. The framework also supports structured results, saved trajectories, detailed logs, and tracing through Arize Phoenix or Langfuse.

## Setup and configuration

Install the package with `uv`. Use `uv tool install mobilerun` for CLI use, or `uv pip install mobilerun` when combining the CLI with Python integration. Python 3.11 through 3.13 is supported; Python 3.14 is not currently supported. Anthropic support is installed separately with the `anthropic` extra.

For Android, install ADB, enable Developer options and USB debugging, then run `mobilerun setup` to install and prepare the Portal app. Run `mobilerun ping` to check the connection. The `mobilerun configure` command guides provider, authentication, and model selection. Provider keys can also be supplied through environment variables such as `OPENAI_API_KEY`, `GOOGLE_API_KEY`, and `ANTHROPIC_API_KEY`. iOS devices use the separate iOS Portal flow.

## Tools and capabilities

The mobilerun MCP server supports:

- Natural-language device commands from the CLI
- Python automation and custom tools
- UI-tree inspection and screenshot-based understanding
- Taps, swipes, text entry, app launching, and device-state access
- Vision and vision-only execution modes
- Reasoning mode for multi-step workflows
- Structured output, app cards, and credential support
- Macro replay, debugging, Docker execution, and telemetry

Models can be supplied through OpenAI, Anthropic, Gemini, xAI, Ollama, DeepSeek, OpenRouter, and OpenAI-compatible providers, subject to the relevant configuration and credentials.

## Limitations and notes

The mobilerun MCP server requires device-specific preparation rather than providing an independent simulator. Android depends on ADB and an accessible Portal installation. iOS setup is handled through a different Portal process. The local framework runs the agent on your machine; managed phones, hosted devices, and cloud API workflows belong to Mobilerun Cloud instead.

For applications without useful accessibility data, screenshot-based operation may be necessary. Longer or more complex tasks may need reasoning mode, and execution can be inspected with debug output, saved trajectories, or supported tracing systems.

_Full upstream README: https://allmcps.com/mcp/mobilerun/readme_