# agent-device

> agent-device is an open-source MCP server for developer tools. Let coding agents drive iOS, Android, TV, web and desktop apps and capture evidence. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, iOS, Android, macOS, Linux, Web, self-hosted
- **Repository:** https://github.com/callstack/agent-device
- **MCP Registry name:** io.github.callstack/agent-device
- **Categories:** [developer tools](https://ai-powered.com/categories/developer-tools), [browser automation](https://ai-powered.com/categories/browser-automation)
- **Listing updated:** 2026-11-18

## About

agent-device, from Callstack, is a device layer for coding agents that need to check their work in a running app. Instead of reasoning over screenshots alone, the agent reads accessibility snapshots, acts through element refs and selectors, and saves screenshots, video, logs, traces and React profiles as evidence. Sessions are scoped to the caller's git worktree, so parallel agents can share devices.

The README lists iOS, Android and HarmonyOS simulators, emulators and physical devices, plus tvOS, Android TV, Fire OS, Vega OS, web, macOS and Linux. Support depth varies by target, and a capabilities command shows what each one supports. Working flows can be saved as .ad replay scripts for CI, or exported as Maestro YAML, and remote device clouds such as BrowserStack and AWS Device Farm can be connected.

It runs as a CLI, a typed Node.js client, or a stdio MCP server started with agent-device mcp. Node.js 22.12 or newer is required, and web automation needs Node.js 24.

## Features

- **Accessibility snapshots:** Reads the app's UI tree with refs the agent can act on.
- **Many targets:** Mobile, TV, web, macOS and Linux apps on simulators and real devices.
- **Evidence capture:** Saves screenshots, video, logs, traces and network data for review.
- **Replay scripts:** Saves working steps as .ad scripts for CI or exports Maestro YAML.
- **React Native profiling:** Looks for renders that slow an interaction.
- **Worktree-scoped sessions:** Coordinates device access across parallel agent worktrees.
- **Remote device clouds:** Connects to BrowserStack, AWS Device Farm, TestMu AI and Limrun.

## Good for

- Have an agent run a new screen on iOS and Android and attach screenshots
- Reproduce a crash and collect the logs leading up to it
- Check a change for needless React Native re-renders
- Turn an explored checkout flow into a CI replay script
- Verify a pull request on a physical device

## Install

### Claude Code (Terminal)

```sh
claude mcp add agent-device -- npx -y agent-device
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "agent-device": {
      "command": "npx",
      "args": [
        "-y",
        "agent-device"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "agent-device": {
      "command": "npx",
      "args": [
        "-y",
        "agent-device"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "agent-device": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "agent-device"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "agent-device": {
      "command": "npx",
      "args": [
        "-y",
        "agent-device"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "agent-device": {
      "command": "npx",
      "args": [
        "-y",
        "agent-device"
      ]
    }
  }
}
```

## FAQ

### Which agents does it work with?

Claude Code, Codex, Cursor, Windsurf, Cline, Goose and others that can run a CLI or connect over MCP.

### What are the requirements?

Node.js 22.12 or newer; web automation needs Node.js 24 or newer.

### How do I start the MCP server?

Run agent-device mcp over stdio.

### Does every platform support every command?

No. Support depth varies, and agent-device capabilities shows what a target supports.

### Is agent-device free?

Yes. agent-device is open source, so you can use and self-host it for free.

### How do I add agent-device to Claude Code?

Run this command in your terminal: claude mcp add agent-device -- npx -y agent-device

### What are the best alternatives to agent-device?

The closest alternatives to agent-device are REA, Browser Use, Scrapling MCP Server, Context7 and Chrome DevTools MCP.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Browser Use](https://ai-powered.com/mcp/browser-use): Browser Use is a freemium MCP server for browser automation.
- [Scrapling MCP Server](https://ai-powered.com/mcp/scrapling-mcp-server): Scrapling MCP Server is an open-source MCP server for web search and scraping.
- [Context7](https://ai-powered.com/mcp/context7): Context7 is a freemium MCP server for developer tools.
- [Chrome DevTools MCP](https://ai-powered.com/mcp/chrome-devtools-mcp-2): Chrome DevTools MCP is an open-source MCP server for browser automation.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/agent-device
