Turn a Spare Mac Into a Throwaway Machine Claude Code Fully Controls

A hands-on guide walks through converting an old Mac into an always-on box that Claude Code can operate autonomously, reachable over SSH from your main machine or by voice through the Claude mobile app. The core idea is isolation: rather than running Claude Code with —dangerously-skip-permissions on a primary machine, you dedicate a sacrificial Mac wiped of personal data and never signed into an Apple ID, so an agent with broad access has nothing sensitive to reach. The author argues this beats containers because a real Mac exposes the full, latest Claude Code features and native apps like Unity that agents can drive via computer use — clicking and dragging — which containers can’t provide, and because a container’s network traffic still routes through your main machine.

The setup steps trade convenience for security by design. You erase the target, create a local admin account, enable Remote Login, and — notably — install a sudoers rule granting fully passwordless sudo so the agent never hits a password prompt. Additional tweaks disable system sleep, clamshell sleep, display sleep, and the screen saver so the machine stays on the network and never locks. SSH key auth is set up from the source Mac, addressed via the stable .local hostname rather than a changeable IP, and a clip.sh helper pipes pbcopy/pbpaste over SSH to move text and images between machines peer-to-peer.

The practical appeal is real, but the security tradeoffs deserve scrutiny: passwordless sudo plus an agent running with skipped permissions means anything that compromises Claude Code — or any prompt-injection payload it ingests during a research task — gets unattended root on a networked machine sitting on your home Wi-Fi. The ‘nothing to lose’ framing holds only as long as the box stays genuinely isolated from the rest of your network and accounts.