Monitor every coding agent on your Mac at once

What there is to know, and where it lives

Every agent CLI is an ordinary process: your process list already knows which agents are running, in which directory, under which terminal. Claude Code goes further — it writes a local transcript (~/.claude/projects/…) as it works, which carries the session title, the last exchange, and what tool it’s touching right now.

The states that matter

• Working — burning CPU (the agent or its build subprocesses) or actively writing its transcript.
• Idle — finished or waiting for your next message. Not an emergency.
• Asking — a permission prompt is up. The only state that deserves a ping.

How NapCat shows it

NapCat samples the process list every few seconds and reads Claude Code’s local transcripts, read-only. Every session becomes a live card in the notch — task, last exchange, model, terminal — with a bar that says Working… in blue while anything runs. Click a card and it lands on the session’s exact terminal tab in Terminal, iTerm2, or Warp.