Mobster — Claude Code MCP server: what it does, how to install
Mobster lets an agent inspect an iPhone screen and send device actions through WebDriverAgent. Use it for supervised mobile workflows and UI checks.
Open on GitHubWhat Mobster does
Mobster is an engine for agent-controlled iPhone tasks. It reads the accessibility tree, presents the current screen and task to a model, checks the proposed action, then taps, types, or swipes through WebDriverAgent. The loop continues until the task proof appears or a run limit stops it.
Its safety layer records events, keeps one run lock per phone, masks passwords and card data, and blocks duplicate actions when an outcome is unclear. It also asks for approval before a send, purchase, post, or delete. Phone text is treated as data rather than as an instruction to the model.
The example in examples/reference_loop.py needs a WebDriverAgent runner serving the iPhone and a Claude or OpenAI key. Without that runner, the example cannot start. Backend tests use make venv and make test-backend; they run without a phone, simulator, network, or key.
What it helps with
- inspect an iPhone screen through its accessibility tree
- carry out taps, text entry, or swipes from an agent task
- ask for approval before a risky device action
- prevent a repeated action after an unclear result
Example prompts for the agent
- Inspect the current iPhone screen and list the controls available to tap.
- Open Settings, read the iOS version, and verify the result on screen.
- Fill in this iPhone field, but ask me before sending it.
How to install Mobster
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Mobster
How do I install Mobster in Claude Code?
In Claude Code, connect Mobster as an MCP server. For iPhone workflows, prepare a WebDriverAgent runner and a Claude or OpenAI key.
What does Mobster do?
Mobster lets an agent read an iPhone screen and perform taps, typing, and swipes. It checks the result after each action and requests approval for sensitive operations.
What is required to run Mobster with an iPhone?
The examples/reference_loop.py workflow requires a WebDriverAgent runner serving the phone plus a Claude or OpenAI key. The backend test suite runs without those runtime pieces.
