The best ways to track Claude Code (and Codex) usage

The short answer

No single tool is “best” for everyone. The right one depends on which question you’re asking: “how much is left before I’m blocked?” or “how many tokens did I burn this month?” Those sound alike but need different data. Let’s untangle that first, slowly, the sloth way.

Two kinds of number: official limit vs local estimate

Every tracker on this page shows one (or both) of these:

1. The official limit percentage

This is what Anthropic or OpenAI reports for your plan: for Claude, how much of your five-hour session and weekly windows you’ve used, plus when each resets. It covers everything that draws on the plan: Claude Code, claude.ai and Claude Desktop, on every device. It’s the number that decides whether you get the “you’ve hit your limit” message.

2. A local token or cost estimate

Claude Code and Codex keep session logs on your machine. Tools can add up the tokens in those logs and multiply by list prices to get a dollar figure. That’s genuinely useful for spotting a heavy day or comparing models. But it isn’t your limit:

• Anthropic doesn’t publish a fixed token budget per plan. Its help center says usage depends on conversation length and complexity, features, model and effort.
• Local logs only see Claude Code on this machine, not claude.ai, Desktop or your other laptop.
• On Pro or Max, the dollar figure is what the tokens would cost at API list price, not your bill. Anthropic’s cost docs say the session cost in /usage “isn’t relevant for billing purposes” for subscribers.

So when a tool shows “72% of your 5-hour block”, check where that 72% comes from. If it’s from Anthropic’s data, great. If it’s tokens divided by a guessed budget, treat it as a rough hint.

Built-in: /usage and the web pages

Free, official and already installed. Start here.

Claude Code: /usage (also /cost and /stats)

The commands reference describes /usage as showing “session cost, plan usage limits, and activity stats”. /cost and /stats are aliases. On a Pro, Max, Team or Enterprise plan you get plan usage bars with reset times, plus a breakdown of what’s eating your allowance: skills, subagents, plugins, MCP servers, long context, cache misses. Press d or w to switch between the last 24 hours and 7 days. Anthropic notes that this breakdown is approximate and computed from local session history, so it doesn’t include claude.ai or other devices. The usage bars themselves come from your account.

Handy details: /usage runs immediately even while Claude is mid-answer, and if the usage request fails it shows your last-known bars from the past hour with a note saying how old they are.

claude.ai: Settings → Usage

claude.ai/settings/usage shows your limits and reset times across all Claude surfaces, plus usage-credit settings. It’s the best place to check when Claude Code and claude.ai seem to disagree. Team and Enterprise admins also have org analytics and spend reports.

Codex: the dashboard, /status and /usage

OpenAI points you to the usage dashboard at chatgpt.com/codex/settings/usage for current limits and reset times. In the CLI, /status shows remaining limits for the active session, and /usage shows token activity and earned resets. Our guide on how to check Codex usage walks through each one.

The catch: all of these answer only when you ask, and only for one tool. That’s fine if you check now and then. It’s easy to forget mid-flow, which is usually right before the limit message turns up.

Status line scripts

Claude Code’s status line runs a command you choose and passes it session data as JSON. For Pro and Max subscribers that JSON includes a rate_limits object with five_hour and seven_day used percentages and resets_at times. Those are Anthropic’s numbers, not estimates. A few lines of shell turn them into something like 5h 42% · wk 61% · resets 15:45 at the bottom of every session.

Things to know from the docs: rate_limits appears only for subscribers and only after the first response in a session, and each window can be missing on its own. Our status line guide has a copy-paste script, and the status line generator builds one for you. You can also run /statusline in Claude Code and describe what you want.

Codex has its own: run /statusline in the Codex CLI and turn on the rate-limits item. The two tools don’t share a line, though, so each terminal shows only its own numbers.

Good for: people who live in the terminal and mostly use one tool. Less good for: seeing your limit when no session is open, or getting a heads-up before you run out.

ccusage and other open-source CLIs

ccusage

ccusage is the best-known open-source tool here, and deservedly so. It’s MIT-licensed, needs no install (npx ccusage@latest or bunx ccusage), and reads the local session logs that coding CLIs leave on disk. Its docs list ~/.config/claude/projects/ and ~/.claude/ for Claude Code and ${CODEX_HOME:-~/.codex} for Codex.

What you get:

• daily, weekly, monthly and session reports of tokens and costs;
• blocks, which groups Claude Code usage into five-hour windows;
• ccusage codex daily for Codex, plus support for several other agent CLIs;
• a statusline command (beta) you can plug into Claude Code’s status line, showing session, daily and block cost, burn rate and context use.

Its own docs are upfront that “costs are estimates and may not reflect actual billing”, calculated from token counts and model pricing. It also says all analysis happens on your machine. What it doesn’t do is read your official plan percentage, so a ccusage block and /usage can legitimately show different pictures.

Great for: cost history, model comparisons, “what did that refactor cost at API prices?”, and anyone on an API key, where token cost is the bill (still check the Console for the authoritative number).

Claude Code Usage Monitor

Claude Code Usage Monitor is a Python terminal dashboard (uv tool install claude-monitor, then claude-monitor), MIT-licensed, with install steps for Linux, macOS and Windows. It reads local Claude Code data and adds burn-rate charts and predictions of when a session runs out. Its README is candid about where numbers come from: it can capture Claude Code’s official rate_limits through an optional status line hook and prefers those when fresh, and otherwise labels figures as local estimates. Its built-in plan presets are its own approximations, not Anthropic figures.

Both tools are free, transparent and well loved. If you want numbers in a terminal and you’re comfortable with “estimate” meaning estimate, they’re excellent.

Menu bar and tray apps

These keep your usage visible all the time, outside any terminal, and usually add notifications. Most get the official limit percentage, by reusing an existing sign-in or a browser session, rather than estimating from logs. That’s more accurate, and it also means you should look at how each app signs in before installing it. A few we looked at, described from their own READMEs:

CodexBar

CodexBar by Peter Steinberger is open source (MIT) and covers a very wide range of providers, Claude and Codex among them. It shows session, weekly and monthly windows with countdowns to the next reset. There’s a macOS menu bar app (macOS 14+, via GitHub Releases or brew install --cask codexbar), a Linux desktop app and a codexbar CLI. Depending on the provider it reuses OAuth, API keys, local files or, opt-in, browser cookies. Its README notes it reads only known locations for enabled features and stores provider tokens in its config file with restrictive permissions. If you juggle many AI tools, it’s the broadest option we found.

Claude Usage Tracker

Claude Usage Tracker is a native Swift menu bar app (macOS 14+, MIT, installable with Homebrew or as a signed, notarized download). It shows the five-hour session, weekly usage, per-model usage and reset timers, and its README says it has tracked Codex limits since version 3.3.0. It can reuse your Claude Code login, sign in through an embedded browser, or take a pasted session key, and stores credentials in the macOS Keychain.

Plenty of others

GitHub has many smaller projects in this space, and new ones appear every month. They differ in maturity, in whether they’re signed and notarized, and in how they authenticate: some read the Claude Code sign-in from the Keychain, some read only Claude Code’s own status line feed, some use a claude.ai cookie. None of them, Quotchi included, is made or endorsed by Anthropic or OpenAI. Before you install one, check three things: how it signs in, whether it’s notarized, and whether it sends anything anywhere.

Quotchi (our app)

Quotchi is the calm Mac menu bar app we make, with a sloth who does the worrying for you. Here’s what it does, straight from the help center:

• Official windows, side by side. Claude Code and Codex usage and reset times in the menu bar, read through each provider’s existing connection. Optional Gemini app usage, without reset times. Older readings stay labeled as stale instead of showing an old number as current.
• Heads-ups. A gentle notification at 80% (or levels you choose), when a limit is reached, and “ping me when it’s back” at reset. An optional estimate like “~3:40 PM” when your recent pace says you’ll run out soon.
• Where you type. An opt-in status line segment such as Claude 5h 42% · wk 61% · Codex 30% that chains your existing status line and can restore it exactly. A quotchi terminal command (quotchi status, quotchi can-i, quotchi wait claude && claude -c).
• Widgets and local Stats (your week, tokens and models), Free.
• Pro adds history, “which session needs you”, a desktop sloth companion and more. It’s a one-time purchase with a 7-day free preview in the app; see pricing.

On privacy: no account, it never reads your prompts, snapshots and history stay on your Mac, and no app analytics are uploaded. The privacy page lists exactly what’s read and sent. It needs an Apple silicon Mac on macOS 14 or later.

Comparison table

Based on each tool’s own docs as of October 17, 2026. “Official %” means the limit percentage reported by Anthropic or OpenAI; “local estimate” means tokens and costs computed from log files.

Which one should you use?

“I just want to know if I can start something big.”

Run /usage. If you do that more than a few times a day, put the numbers in your status line so you never have to ask.

“I want to know what my usage would cost, or which model burns the most.”

ccusage. It’s the clearest token and cost history around. Remember the dollars are list-price estimates; on a subscription, our API value calculator is a fun way to see what your plan is worth at API prices.

“I want burn-rate charts in a terminal.”

Claude Code Usage Monitor, ideally with its status line hook so it can use official numbers.

“I use lots of AI providers, or I’m on Linux.”

CodexBar covers the most providers and has a Linux app.

“I use Claude Code and Codex on a Mac and want a calm glance plus a ping when I’m close or a limit is back.”

That’s exactly what we built Quotchi for. Claude Usage Tracker and CodexBar are good alternatives worth comparing.

When Quotchi isn’t the right choice

We’d rather you pick the right tool than the wrong one with a sloth on it. Skip Quotchi if:

• You’re on Windows, Linux or an Intel Mac. Quotchi needs Apple silicon and macOS 14+. Look at CodexBar (Linux), Claude Code Usage Monitor, or a status line script.
• You want detailed token and cost accounting. Quotchi’s Stats are a friendly weekly summary, not a ledger. ccusage is the better fit.
• You use Claude Code through an API key, a cloud provider (Amazon Bedrock, Google Cloud, Microsoft Foundry) or a gateway. There’s no plan window to show; your spend lives in the Console or your cloud billing, and Anthropic’s docs recommend OpenTelemetry for per-user metrics.
• You track many providers beyond Claude Code, Codex and Gemini. CodexBar supports far more.
• You check once a day and the status line already covers it. Honestly, you don’t need another app. Enjoy the free script.
• You need fully open-source software. Several alternatives above are MIT-licensed; Quotchi isn’t open source.

Common mistakes and myths

“My token count tells me exactly how close I am to the limit.”

Anthropic doesn’t publish a fixed token budget, and the allowance is shared with claude.ai and Desktop. Tokens are a hint, the official percentage is the answer.

“The cost figure is what I’m paying.”

On Pro or Max it’s an API list-price estimate. Your bill is your subscription plus any usage credits.

“Switching trackers will fix a number that looks wrong.”

Usually the account is the issue. Check /status to confirm you’re on your subscription, and that ANTHROPIC_API_KEY isn’t set.

“Menu bar apps are official.”

None covered here is made by Anthropic or OpenAI, including ours. Check how each one signs in.

What changed recently

Sources

Pages we read on October 17, 2026. If anything here disagrees with them, they win.

• Claude Code docs: Commands
• Claude Code docs: Manage costs effectively
• Claude Code docs: Status line
• Claude Help Center: How do usage and length limits work?
• Codex docs: Pricing and usage limits
• Codex docs: Slash commands in Codex CLI
• ccusage on GitHub and ccusage guide (statusline)
• Claude Code Usage Monitor on GitHub
• CodexBar on GitHub
• Claude Usage Tracker on GitHub
• Quotchi help: Using Quotchi and Quotchi privacy

FAQ