GUIDE
Run Claude Code overnight on your Mac
Yes, you can hand Claude Code a long task at midnight and read the results with coffee - but only if three things hold through the night: the Mac stays awake, the screen stays private, and something gathers up what happened so you don't scroll a dozen terminals at 7am.
The problem with walking away
A long autonomous run is the best reason to own a coding agent and the easiest thing to sabotage. You queue the work, close the lid of your attention, and come back to a login window, a stalled agent, and no idea which of your four sessions actually finished. Getting an overnight run right is really three separate jobs.
Job one: keep the Mac awake - all three ways
Sleep is not one setting, it is three, and most keep-awake tools only cover two of them.
- System sleep - a power assertion like
PreventUserIdleSystemSleep(whatcaffeinate -iholds) stops the machine idling out. - Display sleep - a second assertion (
PreventUserIdleDisplaySleep) keeps the screen powered. - The lock timer - the screen-saver / require-password timer is a third mechanism that power assertions never touch. With “require password after screen saver” on, your Mac can lock over a perfectly awake session and freeze the agent behind the login window. The fix is a heartbeat that periodically declares user activity (
IOPMAssertionDeclareUserActivity), resetting the lock timer the way a mouse nudge would.
NapCat holds all three automatically whenever a watched agent is working, and lets go the moment the last one stops. There is an optional battery guard too: release below 15% unplugged, so a forgotten overnight run can’t drain the Mac flat. The full mechanics are in keep your Mac awake while agents run.
Job two: a screen that stays private while it works
A screen you leave on all night is a screen anyone can read. NapCat’s nap overlay is the honest middle ground: one shortcut (⌥⌘N) drops a frosted “asleep” screen over every display - your code, prompts and file tree hidden - while Claude Code keeps running underneath and keep-awake stays engaged. It looks asleep. It is very much not asleep.
Job three: a Morning Report, not a scavenger hunt
The overnight run only pays off if the morning is fast. Wake the nap screen after a long stretch and NapCat has a briefing ready before the frost even lifts: one aggregated debrief for the whole hive - what finished, what is blocked, what changed, and what needs you first - instead of a stack of per-session cards you have to reconstruct. You read the one thing to do next, not twelve transcripts.
Underneath it, each session is still a live card: task, last exchange, model and terminal, with a blue Working… bar while anything runs and an Asking state the second a permission prompt goes up. Click a card and it lands on that session’s exact terminal tab in Terminal, iTerm2 or Warp.
The overnight checklist
- Queue your task and confirm the session shows as Working.
- Hit ⌥⌘N to nap - keep-awake and privacy engage together.
- Leave it plugged in (or trust the battery guard to bail out safely).
- In the morning, wake to the Morning Report and start with whatever it says is blocked or waiting on your yes.
If you also want to watch your rate limits across a long night, see how to check Claude Code usage, and the full feature list lives on the Claude Code page.
Questions
Will my Mac lock and stall the run even if it never sleeps?
It can, and this is the failure most keep-awake tools miss. The screen-lock timer is separate from system and display sleep. NapCat runs an activity heartbeat that resets the lock timer while an agent works, so the login window never freezes your run.
Do I need Claude's paid Remote Control or a Max subscription to approve prompts overnight?
No. NapCat's notch approvals work locally through Claude Code's Notification hook - you get Allow and Deny buttons in the menu bar without Anthropic's Remote Control feature or a Max plan. It's a separate, local path to the same yes-or-no.
Is it safe to leave it running on battery?
Turn on the optional battery guard. It releases keep-awake below 15% while unplugged, so an overnight run can't flatten the Mac. On the charger, it simply runs until the last agent stops.
What do I actually see when I wake up?
The Morning Report: one aggregated briefing - what finished, what's blocked, what changed, what needs you first - already composed before the nap screen's frost lifts, plus live per-session cards you can click straight to their terminal tab.
Does NapCat send my overnight work anywhere?
No telemetry, no account. NapCat reads your process list and Claude Code's local transcripts read-only. The one optional network call is your own Claude usage meter, using your own login.
NapCat does all of this from your Mac’s notch - for five dollars, once.
Download - free 2-day trial