Use Mobster from Claude Code

Mobster adds its MCP server to Claude Code for every project on this Mac. Claude Code then lists Mobster’s tools, including verify_start, screen, tap and run_tests. See the MCP server for the full list.

Add it

mobster mcp install claude-code

It prints what it changed:

Mobster's MCP server: /Applications/Mobster.app/Contents/MacOS/mobster-agent mcp
  Claude Code     added      ~/.claude.json
New sessions of the other clients pick it up. Check it with `mobster mcp doctor`.

Start a new Claude Code session to load it. Add --dry-run to see the change first and write nothing. Run the command again and nothing changes. Without the mobster command on your PATH, run Claude Code’s own command with the agent inside Mobster for Mac:

claude mcp add --scope user mobster -- /Applications/Mobster.app/Contents/MacOS/mobster-agent mcp

--scope user adds it to every project. Without it, Claude Code adds the server to the current folder only. Mobster for Mac does the same from Settings › Connectors: Add beside Claude Code, under Use Mobster from….

What it writes

When the claude command is on your PATH, Mobster runs claude mcp add --scope user, as above. Otherwise it writes the same entry in ~/.claude.json, or in $CLAUDE_CONFIG_DIR/.claude.json when that is set:

~/.claude.json

{
  "mcpServers": {
    "mobster": {
      "type": "stdio",
      "command": "/Applications/Mobster.app/Contents/MacOS/mobster-agent",
      "args": ["mcp"],
      "env": {}
    }
  }
}

It writes the full path to mobster, and it never writes a key. --project DIR writes DIR/.mcp.json instead, a file you commit so everyone on the project gets the server.

Check it

mobster mcp doctor

The doctor starts the server the way Claude Code does, lists its tools and calls status. It also checks that Claude Code’s entry points to an existing mobster executable. On 26 Oct 2026 (Mobster 0.4.0, no key set) it printed:

Server: /Applications/Mobster.app/Contents/MacOS/mobster-agent mcp
  started with launchd's PATH (/usr/bin:/bin:/usr/sbin:/sbin), as an app from the Dock would start it
  initialize             659 ms
  tools/list              14 ms
  tools/call status      799 ms
  130 tools: status, verify_start, wait, screen, tap, type_text, swipe, ...
  status: Mobster 0.4.0. Smart: off (no OpenAI or Anthropic key; set OPENAI_API_KEY or ANTHROPIC_API_KEY, or pass --env-file).
  ...
OK in 1549 ms.
  Claude Code     set up     ~/.claude.json

In Claude Code, /mcp lists mobster.

Check a screen

Build your app for the simulator, then ask Claude Code in your project’s folder:

Use Mobster to check that the app's first screen shows "Today".

Claude calls verify_start with the app’s path and the expectation, drives the simulator with screen and tap, and calls verify_finish. verify_finish checks the declared expectations. You get a verdict of passed, failed, needs_review or couldnt_run. You get the report path and an overlay frame when available. No model grades the outcome. Claude drives with its own model, so Mobster needs no key. The key-less loop

Add the skill

mobster mcp install claude-code --with-skill

This copies the skill into ~/.claude/skills/ios-testing, so Claude also knows the verify loop and the rules for a real iPhone: ask before anything that sends, buys, posts or deletes, and never enter a passcode. The Claude Code plugin adds the skill too. --with-xcode-tools also adds Xcode’s own MCP tools (xcrun mcpbridge, Xcode 26.3 or later) to Claude Code.

Remove the MCP server

mobster mcp install claude-code --remove

This removes the mobster entry and leaves your other servers alone. Claude Code’s own reference for MCP servers is at https://code.claude.com/docs/en/mcp.

Your coding agent

Apps and tools

From anywhere

In the terminal

Use Mobster from Claude Code

Add it

What it writes

Check it

Check a screen

Add the skill

Remove the MCP server

​Add it

​What it writes

​Check it

​Check a screen

​Add the skill

​Remove the MCP server

Add it

What it writes

Check it

Check a screen

Add the skill

Remove the MCP server