Observed arrival · 2026-11-14
phonedriver puts an AI agent behind a phone screen
An MCP server that lets coding agents read and operate an Android phone or iPhone through screen text, taps, and typing, with approval before payment.
- For
- Developers connecting AI agents to phones
- Worth noticing
- Its recorded Domino’s run shows ambiguous “COLLECT” matches and a missed text entry before the agent recovers.
Field notes
Setup is exposed through npm commands: `npx -y phonedriver doctor` checks the environment, while `npx -y phonedriver mcp` starts the MCP server; the page also supplies client-specific snippets for Claude Code, Codex, and Cursor. A recorded Domino’s ordering attempt shows the agent handling cookie preferences and screen changes, but also encountering duplicate “COLLECT” matches and text that did not land in the intended field. The page reports a nine-job token comparison, which the extracted evidence does not independently verify.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue