Compared
Claude Code usage monitors for Mac, compared
“Usage monitor” covers two different tools: a meter that keeps how much of your plan is left on screen while you work, and a spend report that tells you afterwards what the tokens cost. Pick by that, then by where the number comes from and what you run on. Most of these are free, and this page says so. CrewTower is the paid one, and the only reason to pick it is the question it adds: which of your sessions is waiting for you.
The two kinds
| Spend report | Meter | |
|---|---|---|
| Answers | What did this cost me? | How much have I got left? |
| When you look | Afterwards, deliberately | While working, without looking |
| Usually reads | Your local session files | Your provider's own record |
| Shape | A command or a dashboard | Something always on screen |
| Examples | ccusage, CodeBurn | CodexBar, ClaudeBar, a notch app |
Several tools below do both, and the line between them is where the number comes from. A figure rebuilt from local files is accurate about tokens, because the tokens are in the files. It is an estimate of how close you are to being stopped, because your plan's limits live with your provider, and work done on another machine or on claude.ai is not on this disk.
What you already have, for free
Inside a running Claude Code session, /usage shows the session's cost, your plan usage limits and activity stats; /cost and /stats open the same screen. For history across every session, ccusage reads the transcripts Claude Code already writes under ~/.claude/projects, with no account and no API key:
# inside a Claude Code session
/usage
# in any terminal
npx ccusage@latest daily
npx ccusage@latest blocks
npx ccusage@latest statusline # for Claude Code's status line (beta)ccusage's old blocks --live view was removed in its version 18; its docs point to statusline for figures while you work. Between these, most of the question is answered. A separate monitor is worth adding when the problem is that you have to remember to ask.
Nine tools, side by side
| Price | Runs on | Claude Code figure from | History | A session waiting for you | |
|---|---|---|---|---|---|
| CrewTower | $9.99 once (early bird, regularly $14.99) | macOS 15 or later, Apple silicon and Intel | Anthropic's usage endpoint, with Claude Code's own login; Codex from its own rate limit reporting | None; with the beta Tell Claude plugin, the current session's cost so far | Every session, for 26 agents; answered from the notch for 6 |
| CodexBar | Free, MIT | macOS 14 or later; Linux app and command line | Claude's OAuth usage API first, then the CLI or the web | Recorded spend by rolling days, month to date or all time | Lists live sessions as active or idle; a waiting state is a non-goal |
| ClaudeBar | Free, Apache 2.0 | macOS 15 or later | claude /usage or the OAuth usage API | 30 days; cost estimated at list prices | An opt-in notch shows “Needs you” for the latest Claude Code session; answering not published |
| Claude Usage Tracker | Free, MIT | macOS 14 or later | Claude Code's credentials or a claude.ai sign-in | Usage history charts | A beta notch HUD pulses when Claude is waiting; it says it never answers for you |
| Usage4Claude | Free, MIT | macOS 13 or later; not notarized | Claude's own undocumented usage API | Under consideration | Not published |
| SessionWatcher | From $6.99 once until 31 October 2026, then $8.99; 7-day trial | macOS 14 or later, Windows | The login already on your computer; some providers' own APIs | 26 weeks kept; cost estimated from tokens | Not published |
| Claude Code Usage Monitor | Free, MIT | Any system with Python 3.9 or later | A local estimate by default; Claude Code's own figures with --statusline | Opt-in local usage warehouse | Not published |
| ccusage | Free, MIT | Any system, from the command line | Local session logs | Daily, weekly, monthly, session and 5-hour block reports | Not published |
| CodeBurn | Free, MIT | macOS, Windows, Linux | Local session files for spend; codeburn quota asks the provider | Spend by tool, model, project and task | Not published |
“Not published” means that tool's own pages say nothing about it, not that it cannot. Pulse for Claude Code, a free local dashboard with a notch strip, goes further than any of these: it answers Claude Code's permission prompts through a PreToolUse hook, on its dashboard or on your phone through ntfy, and falls back to the terminal prompt after 60 seconds.
Where they are better than CrewTower
Almost all of them are free. Seven of the eight are free and open source; SessionWatcher is the other paid one.
Most keep a history. CodexBar, ClaudeBar, Claude Usage Tracker, SessionWatcher, ccusage and CodeBurn all keep spend or usage over time. CrewTower keeps none.
More providers. CodexBar lists 106, ClaudeBar 28 and SessionWatcher 20 tools. CrewTower's meter covers Claude Code and Codex.
Windows and Linux. SessionWatcher and CodeBurn have Windows builds, CodeBurn and CodexBar run on Linux, and ccusage and Claude Code Usage Monitor run anywhere. CrewTower is a Mac app.
Older Macs. Usage4Claude runs on macOS 13, and CodexBar, Claude Usage Tracker and SessionWatcher on macOS 14. CrewTower needs macOS 15.
Where CrewTower is different
The provider's own figure, against the estimators. ccusage, CodeBurn's spend reports and Claude Code Usage Monitor's default view rebuild usage from local files. CrewTower reads Claude Code's figure from Anthropic's usage endpoint with the login Claude Code already uses, about every five minutes, so it is the provider's own figure rather than a reconstruction; Codex's comes from Codex's own rate limit reporting. CodexBar, ClaudeBar and Claude Usage Tracker read the provider's figure too, so this is no advantage over them.
The waiting session, answered. The usage strip is one line under the sessions. The rest of CrewTower shows every running agent session in the notch and brings forward the one that has stopped for you. For Claude Code, Codex, Cursor, Gemini CLI, Qwen Code and opencode you allow or deny the request from the card; the agent's hook is waiting on a local socket and the answer goes back down it, so the terminal never comes forward. When you do want the terminal, Jump lands on the exact tab or tmux pane.
Context, next to usage. Each card shows how much context the session has left, for the agents that report it, and for Claude Code it suggests compacting at a task break before quality starts to slip.
The usage meter is optional: on by default, switchable off, and one of only three things in the app that make a network request. The privacy page lists all three and names the exact credential and URL.
Using more than one
Nothing here conflicts. A common setup is a spend report for the month (ccusage or CodeBurn), a meter for the providers you use (CodexBar or ClaudeBar), and CrewTower for the sessions. If you already keep a meter in the menu bar, CrewTower's own usage strip can be switched off.
Which one to pick
What did I spend? ccusage on the command line, CodeBurn for a desktop view by project, or SessionWatcher if you want a paid app with a trial.
How much is left, across many providers? CodexBar, or ClaudeBar on macOS 15.
How much is left, for Claude Code alone, free? Claude Usage Tracker or Usage4Claude, or /usage itself.
Which session needs me, and answering it without switching windows? CrewTower, with any of the meters above beside it if you want more than Claude Code and Codex. If you want that answered from the notch but would rather not pay, CodeIsland and Open Island are free and open source; the comparison table has both.
Questions
What is the best Claude Code usage monitor?
It depends on the question. For what you have spent, by day, session and model, ccusage is the standard answer, free and open source. For how much of your window is left while you work, use something always on screen: CodexBar if you use many providers, or a Claude-only meter such as Claude Usage Tracker. CrewTower is the one to add if you also want to know which session is waiting for you.
Claude Code Usage Monitor or ccusage?
Claude Code Usage Monitor is a live terminal view of the current window, estimated from local logs unless you feed it Claude Code's own figures with --statusline. ccusage prints reports of what you spent by day, week, month and session. One is for watching, the other for looking back, and many people run both.
Is there a Claude Code usage monitor for Windows?
Not CrewTower: it is a Mac app and needs macOS 15 or later, on Apple silicon or Intel. SessionWatcher and CodeBurn publish Windows builds, and ccusage and Claude Code Usage Monitor run on Windows from the command line.
How does it decide when to suggest compacting?
It suggests it for Claude Code only, and only at a task break: the turn has ended and Claude is waiting for your next prompt, not stopped on a question. The suggestion comes once the conversation passes about 300k tokens on a 1M window, or 60% of a smaller one (about 120k on 200k), and again after each further 100k. Close to Claude Code's own auto-compact point it says “auto-compact soon” instead.
Sources
Read on 19 October 2026, each from the tool's own site or repository: CodexBar and codexbar.app, with its sessions design notes; ClaudeBar, with its notch notes; Claude Usage Tracker, with its changelog; Usage4Claude; sessionwatcher.com, with its facts page and usage stats page; Claude Code Usage Monitor; ccusage.com, with the live monitoring note, and its repository; CodeBurn; Pulse for Claude Code. Claude Code's commands from its own reference.
All of these are independent projects. Nothing here is a claim about anyone's code; it is what each one publishes, and every cell is checkable in a minute. If a cell is out of date, the tool's own page is the one to trust.
See the window, and the session waiting on you
Explore CrewTower in the interactive website demo. A license gives you access to the Mac app.
Get CrewTower for Mac$9.99 once · macOS 15 or later