Claude Code monitoring tools, compared.

Terminal commands and scripts

The default is to read what the CLI prints: usage commands, scrollback, and the occasional script to total things up. It is free and always available, but it is pull, not push. You only see usage when you stop and ask for it, there is no alert before a limit, and parallel sessions mean parallel terminals to babysit.

AgentPeek makes the same local data ambient: your token usage sits under the notch and updates as you work, with budget alerts so you are warned before you run out.

Terminal multiplexers

A multiplexer like tmux tiles many sessions into one window, which helps you see more at once. But the panes still show raw output, not agent state: there is no notion of which session is waiting on a permission, how much budget each is using, or which dev server belongs to which project.

AgentPeek is agent-aware. Each session row shows live state, recent activity, and usage, and permission prompts surface inline so nothing waits silently.

Generic menu bar monitors

System menu bar apps are great for CPU, memory, and network, and some can show a custom script's output. They are not built for coding agents, though, so you end up scripting the agent integration yourself and still lack inline prompts, plan approvals, and per-project dev server controls.

AgentPeek is purpose-built for Claude Code and Codex, with a notch surface and a local dev server manager included.

Where AgentPeek fits

AgentPeek is the agent-aware option: a native Mac app that puts Claude Code and Codex in the notch and menu bar with live sessions, token usage, inline prompts, and local dev servers. It is local-first, with no accounts, analytics, or telemetry, and it is a $15 one time license after a 2-day free trial. If you only ever run one session and never worry about limits, the terminal is fine. If you run several agents and want to stop context-switching, AgentPeek is built for that.

Frequently asked questions