How to see what Claude Code and Codex are costing you, on a Mac

You can check what Claude Code and Codex cost you in three ways: the commands built into the agents, open-source terminal tools, and menu-bar apps. This post covers each, with what it shows and who it suits. Oye Collective makes Actvt, one of the apps below. Every other tool is described from its own README or site, with the date we read it, and we say where another tool is the better pick.

As we wrote in What your agents do when you're not looking, one prompt can turn into dozens of model calls, subagents and retries, and the agent doesn't report what they cost. On a subscription there are two numbers to know: how close you are to the plan's limits, and what the work would have cost at API rates. Some tools below show the first, some the second, and a few show both.

Start with what's built in

Claude Code: /usage. As Anthropic's cost docs describe it, the Session block shows the current session's tokens by model and a dollar figure computed locally at list price, with a prompt-cache line on recent versions. On a Pro, Max, Team or Enterprise plan the same screen adds plan usage bars and a breakdown of what counted against your limits: skills, subagents, MCP servers, scheduled loops, and behaviours like long context or cache misses. The session totals reset at /clear, and the breakdown covers this machine only. The same plan usage is on the web at claude.ai/settings/usage.

Codex: /status and /usage. In the Codex CLI, /status shows the session's configuration and token usage, and /usage shows your account's token activity by day or week (Codex command reference). Neither shows a dollar figure.

Who this suits: anyone who checks once in a while, uses one agent, and wants an answer about the session in front of them. It costs nothing and it's already installed. It doesn't keep history across sessions, show both agents in one place, or stay on screen while you work. Claude Code also deletes session files older than 30 days by default (cleanupPeriodDays), so every tool that reads those files loses them too, unless you raise the setting.

Terminal tools

ccusage: run npx ccusage@latest and it reads the agents' local logs and prints daily, weekly, monthly and per-session reports, with costs in dollars at API rates, cache tokens and per-model breakdowns. A blocks report follows Claude's 5-hour windows, and there's a beta statusline for Claude Code. It reads 18 agents, including Claude Code, Codex, OpenCode, Amp, Gemini CLI, Copilot CLI and pi. It fetches model pricing over the network by default, and --offline uses cached prices instead. MIT licensed.

Claude Code Usage Monitor (claude-monitor) is a live terminal view of Claude Code usage against your plan: burn rate, pace, and a forecast of when you'll hit the limit. It can read Claude Code's own rate-limit data through the statusline and labels anything else as a local estimate. It installs with uv tool install claude-monitor, runs on macOS, Linux and Windows, and is MIT licensed. It covers Claude Code only.

Who these suit: people who live in a terminal, want numbers they can script (both have JSON output), or use agents beyond Claude Code and Codex. You have to remember to run them, and they report token counts and costs but not the sessions' contents.

Menu-bar apps

CodexBar puts usage limits for more than 100 AI providers in the menu bar: session, weekly and monthly windows with reset countdowns, credit balances, spend where the provider exposes it, local cost scans, and provider incident badges. It's free and MIT licensed, runs on macOS 14 and later, has a Linux desktop app, and ships a codexbar CLI. To read some providers it reuses your existing logins: OAuth, API keys, and (opt-in) browser cookies, as its README explains.

SessionWatcher is a paid, native tracker for about 20 tools, on macOS 14+ and Windows 10/11. It shows 5-hour and weekly limits with reset times, token and dollar breakdowns, a usage-rate chart, and alerts at 70% and 90%. Pro adds several Claude or Codex accounts, 90 days of history and widgets. Prices are one-time, from $6.99 for Claude on one device to $59 for Pro Lifetime (rising to $8.99 and $79 on 1 November 2026), or $24 a year. There's also a teams plan at $40 a seat per year. It says it reads usage statistics only and that they stay on your machine.

Actvt is ours. It's a native app for Apple Silicon Macs on macOS 15.5 or later, and it reads the session files Claude Code, Codex and pi already write to disk. It shows full transcripts you can search and filter, the live state of running sessions, spend over time by project with a daily heatmap and cache hit ratio, and plan usage meters with reset countdowns. It also shows live CPU, GPU, memory, network and listening ports. It scans transcripts for secrets and personal data on-device, and over SSH it can show sessions from other Macs and Linux machines. The free tier covers system monitoring and the last 72 hours of sessions. A one-time $9.99 tier adds remote machines and 30 days. Premium ($7.99 a month billed yearly, or $14.99 monthly) adds unlimited history, the spend dashboard and the plan meters.

Which one to pick

Pick the built-in commands if you check now and then and only use one agent.

Pick ccusage if you want a free, scriptable number across many agents and don't mind running a command.

Pick Claude Code Usage Monitor if you're on a Claude plan and mostly want to know whether you'll run out before the window resets.

Pick CodexBar if you want limits for many providers at a glance, for free, open source, or on Linux.

Pick SessionWatcher if you need Windows, tools beyond Claude Code and Codex, several accounts, or seats for a team.

Pick Actvt if you want to see what your agents did as well as what they cost: reading back transcripts, catching a secret that ended up in one, watching your Mac while agents run, or following sessions on other machines.

These tools can run side by side. A limits tracker and ccusage report different things, and you can use both.

Why we write about this

We build agent systems inside other businesses, and teams running agents ask us what the agents are doing and what they cost. The tools above cover one developer's Mac. Across a company the same questions need traces, evals and audit trails, which is the observability work we set up for clients.

Sources

Read on 15 October 2026. We'll recheck them each quarter and change the date at the top when we do.

• Claude Code: Manage costs effectively (Anthropic docs)
• Codex: command reference (OpenAI docs)
• ccusage: README
• Claude Code Usage Monitor: README
• CodexBar: README
• SessionWatcher: sessionwatcher.com
• Actvt: actvt.io

If something here is out of date, tell us through oyecollective.com and we'll correct it.