Home / Guides / Claude Code Usage Trackers

Claude Code Usage Trackers Compared: Choose by Workflow

There is no single best tracker for every job. The right choice depends on whether you need a quick official check, an in-terminal indicator, historical reports, or system-wide visibility on a Mac or Windows PC.

Last updated November 15, 2026 · Menu bar apps re-read November 15, 2026; Anthropic docs and terminal tools checked October 6, 2026 · By Soren Starck

Claude usage trackers at a glance

Verdict: for Claude alone on a Mac, the free, open-source Claude Usage Tracker does the most. For many providers at once, CodexBar is free and also runs on Linux. For a Mac and a Windows PC on one licence, with a warning before the limit across 22 tools, SessionWatcher (ours, paid). To check once with nothing to install, type /usage in Claude Code.

AppPriceRuns onCoversWarns before the limit
Claude Usage TrackerFree, MITMac, macOS 13+Claude (per-model limits, API Console costs) and CodexAt 75%, 90%, 95% or custom
CodexBarFree, MITMac (macOS 14+), Linux; no Windows buildMore than 100 providers, Claude among themOptional session quota notifications
ClaudeBarFree, Apache-2.0Mac, macOS 15+“Claude, Codex, Gemini, Copilot, Cursor and 15 more”A notification before you run out
ClaudeUsageBarFree, MITMac, macOS 12+ClaudeAt 25%, 50%, 75% and 90%
Usagebar$9 one-timeMacClaude Code, plus session context leftAt 50%, 75%, 90% and low context
SessionWatcher (ours)From $8.99 one-time; 7-day trialmacOS 14+ or Windows 10 (2004+)/1122 tools, Claude Code among themAt 70% and 90% by default, or every 10%, and on reset

From each project's own GitHub page or site, read November 15, 2026; we did not install them. SessionWatcher's row is first-party, from our facts page. Windows apps are in the Windows section.

Answer first

  • Check now, no install: Claude Code's built-in /usage.
  • Keep 5-hour and 7-day limits in view inside the terminal: Claude Code's official custom status line, free and documented for Pro and Max.
  • A free menu bar app on a Mac: Claude Usage Tracker for Claude (and Codex), or CodexBar if you track many providers; both are open source.
  • Paid, Claude only, on a Mac: Usagebar, $9 once, which also warns when a session's context runs low.
  • Free on Windows: CodeZeno's Claude Code Usage Monitor in the taskbar, Usage Monitor for Claude or OpenQuota in the tray.
  • Reports across several agent CLIs: ccusage, which lists 18 CLIs plus JSON output.
  • Several AI tools on a Mac and a Windows PC, one licence, warnings before the limit: SessionWatcher, our own product, which is paid.
  • Reconcile a bill: your account's own billing view, not a locally estimated session cost.

Source policy: every capability claim below is attributed to a primary source and was checked on October 6, 2026, and the Mac menu bar apps again on November 15, 2026: Anthropic's own documentation for Claude Code, and each project's own README or website for the third-party tools. We did not install or test the third-party tools. “Not documented” or “Not stated” means the cited source did not state the capability on that date; it does not prove the capability is absent. SessionWatcher is our own product, so its entries are first-party and come from our facts page, which is read from the app's release configuration.

How to choose

Six questions decide it, roughly in this order:

  • Where should the number live? Inside Claude Code (the status line), in a terminal you open, in the Mac menu bar or Windows taskbar, or in a browser.
  • Which tools and which computers? Claude alone on one Mac is a different problem from Claude, Codex and Cursor on a Mac and a PC.
  • Do you want a warning? Some trackers notify you at a threshold before the limit hits; others only show a number when you look.
  • How many Claude accounts? Work and personal logins need a tracker that shows both.
  • Free or paid? Most good options are free and open source. A paid app has to earn its price with something specific.
  • How does it sign in? Reusing Claude Code's own login is the least work; pasting a claude.ai cookie is more fragile and has to be redone when it expires.

What each number means

Usage tools can put different numbers next to a percent sign or dollar sign. Treating them as interchangeable leads to bad decisions. Four different numbers are in play, each with its own source:

Plan limit used, and reset time

Source: Anthropic, from your subscription. Five-hour and seven-day windows. Availability depends on plan and account type.

Session tokens and context

Source: the current Claude Code session, on this machine. Context fullness is not a plan limit.

Estimated cost

Source: computed locally from token counts at list rates. An estimate, not a charge.

Authoritative billing

Source: your account's own billing view, such as the Claude Console for API usage. This is the figure that settles what you actually pay.

  • Plan-limit usage: the share consumed in a five-hour or seven-day subscription window. Anthropic's status-line documentation defines rate_limits.five_hour.used_percentage, rate_limits.seven_day.used_percentage, and matching resets_at timestamps. It documents those fields as appearing only for Claude.ai Pro and Max subscribers after the first API response in a session, with each window independently able to be absent. Limits themselves vary by plan and account.
  • Context-window usage: how full the current conversation context is. It is not a subscription rate limit, and a context or auto-compact warning is not a usage-limit warning. See our Claude Code rate-limits guide for the distinction.
  • Tokens and recorded activity: local session records can be aggregated into daily, monthly, model, or project reports. These describe activity recorded on that machine. Anthropic notes the same limitation for the /usage breakdown: the figures are approximate and computed from local session history, so usage from other devices or from claude.ai is not included.
  • Estimated session cost: Anthropic says Claude Code computes this dollar figure locally from token counts priced at standard list rates, so it does not reflect promotional pricing or contracted discounts and may differ from your actual bill. Anthropic also states the session cost is not relevant for billing purposes on Max and Pro subscriptions, where usage is included in the plan. See the Anthropic cost documentation.
  • Authoritative billing: for API-billed usage, Anthropic points to the Claude Console Usage page. Subscription, Team, and Enterprise accounts follow different documented paths, so check the cost documentation for the one that matches your account.

Choose by task

Each workflow below states the setup it needs and links to the primary source the capability was read from, so you can re-check it against a newer version yourself.

1. I use one Claude session and need to check now

Run Claude Code's built-in /usage. Anthropic documents a Session block with token statistics and a locally computed dollar figure, intended for API users. On a Pro, Max, Team, or Enterprise plan the same screen documents plan usage bars, activity stats, and a usage breakdown, with d and w switching between the last 24 hours and the last 7 days. Anthropic also documents what happens when the plan-limit request fails: /usage shows the last usage bars loaded on that machine within the past 60 minutes, labeled as last-known usage, and r retries.

Setup: none beyond Claude Code itself. Source: Anthropic, Manage costs effectively (checked October 6, 2026). Our own walkthrough: how to check Claude Code usage.

2. I use Pro or Max and want 5-hour and 7-day limits beside the terminal

Claude Code's custom status line runs a local script and receives JSON session data on stdin. Anthropic documents fields for context-window usage, an estimated session cost in cost.total_cost_usd, and rate_limits percentages and reset timestamps for the five-hour and seven-day windows. The documentation limits those rate-limit fields to Claude.ai Pro and Max subscribers after the first API response in the session, and each window can be absent, so a script should handle missing values. This is free and official, but it is visible only while Claude Code is open, and the script is yours to maintain.

Setup: run /statusline and describe what you want, or add a statusLine command to settings.json and make the script executable. Source: Anthropic, Customize your status line (checked October 6, 2026).

3. I switch between Claude, Codex, and Gemini and want daily or monthly reports

ccusage reads local CLI usage data and prints reports. Its README lists 18 agent CLIs, including Claude Code, Codex, OpenCode, GitHub Copilot CLI, Gemini CLI, Antigravity, Kimi, Qwen and Grok Build CLI (Cursor is not among them), plus a Claude five-hour block report with active monitoring, JSON output, and a beta Claude Code status-line integration. That combination fits when reporting across several local CLI data sources matters more than a graphical interface. Because it reads local data, its reports cover what those CLIs recorded on that machine.

Setup: a terminal, using the install and report commands in the README. Source: ccusage README (checked October 6, 2026; latest release v20.0.26, September 27, 2026). Capability claims here are version-sensitive.

4. I want a Claude-focused live terminal monitor with forecasts and exports

Claude-Code-Usage-Monitor (CCUM) is the closer fit here. Its README documents a live terminal UI, capture of official status-line rate_limits values, labeled local estimates when official values are not fresh, forecasts, state and JSON and CSV output, and an optional local history store. It is Claude-focused rather than multi-agent, and it separates official, estimated, and unknown values explicitly, which is useful when you care where a number came from.

Setup: a terminal install, plus a status-line hook if you want the official rate-limit capture rather than local estimates. Source: Claude-Code-Usage-Monitor README (checked October 6, 2026; latest release v4.0.0, June 27, 2026). Capability claims here are version-sensitive.

5. I want a menu bar or taskbar app, with a notification before the limit

This is where most of the choice is, and most of it is free. An app at the operating-system level shows the limit while Claude Code is closed or behind other windows, and can notify you at a threshold. The table below lines up the ones people ask about, from each project's own README or site.

AppPriceRuns onCoversWarningsSeveral Claude accountsSigns in with
Claude Usage TrackerFree, MITMac, macOS 13+Claude (per-model limits, API Console costs) and CodexAt 75%, 90%, 95% or customUnlimited profilesClaude Code login, in-app sign-in or session key
CodexBarFree, MITMac (macOS 14+), Linux; no Windows buildMore than 100 providers, Claude among themOptional session quota notificationsNot stated for ClaudeOAuth, browser cookies, CLI or API keys, by provider
ClaudeBarFree, Apache-2.0Mac, macOS 15+“Claude, Codex, Gemini, Copilot, Cursor and 15 more”A notification before you run outNot statedThe provider tools on the same Mac
ClaudeUsageBarFree, MITMac, macOS 12+ClaudeAt 25%, 50%, 75% and 90%Not statedA cookie copied from claude.ai
Usagebar$9 one-timeMacClaude Code, plus session context leftAt 50%, 75%, 90% and low contextNot statedClaude Code’s Keychain login
SessionWatcher (ours)From $8.99 one-time; 7-day trialmacOS 14+ or Windows 10 (2004+)/1122 tools, Claude Code among themAt 70% and 90% by default, or every 10%, and on resetSeveral Claude and Codex accounts (Pro)The logins your tools already keep

Third-party rows: each project's README or website, checked November 15, 2026; we did not install or test them, and “Not stated” means the source is silent. SessionWatcher row: first-party, from our facts page. Head-to-head pages: vs CodexBar, vs Usagebar, vs ClaudeUsageBar.

  • Claude Usage Tracker is the strongest free option for Claude alone on a Mac: unlimited profiles shown at once, a terminal launcher per profile, per-model limits, API Console costs and a status-line integration. It sends a version-only analytics heartbeat once a day.
  • CodexBar is the free choice when Claude is one provider among many: more than 100 in its README, a Linux desktop app and a CLI as well as the Mac menu bar. Its author ships no Windows build; community ports such as Win-CodexBar fill that gap.
  • Usagebar is the paid Claude-only option, $9 once, and the one here that warns about the session context window as well as the plan limit.
  • ClaudeUsageBar still runs on macOS 12, the oldest floor here, but needs a cookie copied from claude.ai.

6. I need bill reconciliation for a team or API account

None of the local tools compared here is the billing authority. For API organizations, Anthropic points to the Claude Console Usage page and workspace-level cost reporting; Team and Enterprise plans have their own admin-side spend reporting; and Pro and Max usage is included in the plan unless usage credits are turned on. Compare a plan charge with a plan-side report, not with a locally estimated session cost.

Setup: account access with the relevant billing or admin permissions. Source: Claude Console Usage and Anthropic's cost documentation (checked October 6, 2026). See also how much Claude Code costs.

Buying SessionWatcher itself for several developers? Pro for Teams is $40 per seat per year (or $4 per seat per month, 3-seat minimum), one Pro licence per teammate on one invoice. It covers Macs only for now, and each person sees their own usage on their own machine; there is no central team dashboard.

Comparison matrix

Cells report what the cited source documented on the verification date, not how good a tool is. Rows cover the criteria that actually change the choice: where the number lives, what it covers, what it costs to run, and how much confidence the evidence supports.

Criterion/usageClaude Code status lineClaude ConsoleccusageCCUMSessionWatcher
Interface and locationCommand inside Claude CodeBar at the bottom of Claude CodeBrowser pageTerminal command; beta status lineLive terminal UI; status-line hookMac menu bar or Windows taskbar (first-party claim)
Visible when Claude Code is closedNoNoYes, in a browserYes, when you run itYes, while the monitor runsYes, in the menu bar or taskbar (first-party)
Sources and tools coveredClaude Code, this machineClaude Code, current sessionAnthropic API account18 agent CLIs listed, including Claude Code, Codex, OpenCode, Copilot CLI, Gemini CLI, Kimi, Qwen, Grok Build CLIClaude-focused22 tools on Pro: Claude Code, Codex, Cursor, Copilot, Gemini CLI, Antigravity, opencode, Devin, Grok, Grok Bot, OpenRouter, Fireworks AI, Z.ai, Kimi, Qwen, DeepSeek, Synthetic, TypeSafe AI (Jev), Command Code, Mistral Vibe, Charm Hyper, Muse Code (Muse Wonder) (from the release-pinned product manifest)
Five-hour usage percentagePlan usage bars documented on Pro, Max, Team, EnterpriseDocumented (rate_limits.five_hour.used_percentage), Pro/Max after first API responseNot documented; Console covers API billingFive-hour block report with active monitoring documentedDocumented from official status-line values, or labeled estimatesYes, with pace (first-party)
Reset timeNot documented on the cited pageDocumented (resets_at, Unix epoch seconds)Not documentedNot documented in the source readNot documented in the source readReset time and countdown (first-party)
Seven-day limitPlan usage bars documented; windows vary by planDocumented (rate_limits.seven_day), same conditionsNot documentedNot documented in the source readNot documented in the source readYes (first-party)
Proactive OS notificationsNot documentedNot documentedNot documentedNot documented in the source readNot documented in the source readAt 70% and 90% by default, or every 10%, and on reset (first-party)
Historical reportsLast 24 hours and last 7 days breakdown documentedNot documented; session-scopedYes, account usage reportingDaily, monthly, session, and block reports documentedOptional local history store documentedDaily usage over 7, 30 or 90 days; tokens and estimated cost (first-party)
Export and automationNot documentedScript receives JSON on stdin; you format the outputConsole reporting and admin APIs documentedJSON output documentedState, JSON, and CSV output documentedNo export
Authoritative for billingNo; cost is computed locally at list ratesNo; same local estimateYes, for API billingNo; derived from local recordsNo; derived from local recordsNo; a local view, not a billing system
PlatformWherever Claude Code runsWherever Claude Code runs; Windows setup documented separatelyAny browserTerminal; see project docsTerminal; see project docsmacOS & Windows
Install and maintenanceNoneYou create and maintain a scriptNone; account accessTerminal install per README; update for new releasesTerminal install per README; optional status-line hookApp install and updates
Data pathLocal session history, plus a request to Anthropic for plan limitsLocal session JSON to your scriptYour account data, in Anthropic's ConsoleReads local CLI usage filesReads local data; optional local storeReads local logins and logs; some providers are queried from your computer; licence checks send the key and device details (first-party)
Price and licenseIncluded with Claude CodeIncluded with Claude CodeIncluded; usage billed to the accountFree open-source project; see repository for licenseFree open-source project; see repository for licensePaid, from $8.99 one-time; 7-day trial that unlocks all of Pro (pricing)
Evidence typeAnthropic documentationAnthropic documentationAnthropic documentationProject README, version-sensitiveProject README, version-sensitiveFirst-party facts page, read from the release configuration

Matrix sources, all checked October 6, 2026: Anthropic cost docs, Anthropic status-line docs, Claude Console Usage, ccusage README (latest release v20.0.26, September 27, 2026), and the Claude-Code-Usage-Monitor README (latest release v4.0.0, June 27, 2026). Open-source capabilities can change between releases, so re-check the README before relying on a row. SessionWatcher cells are first-party because we publish this guide; they come from our facts page.

Where we fit

SessionWatcher: what the money buys

SessionWatcher is our own product, so treat this as a disclosed first-party description, not a review. It is paid, and several free apps above do part of what it does. CodexBar covers more providers; Claude Usage Tracker does more for Claude alone. The differences that are real:

  • A native Windows app. The readout sits inside the Windows 11 taskbar, beside the clock (a small window next to it on Windows 10), on x64 and ARM64.
  • One key for Mac and Windows. Pro covers up to 5 devices in any mix.
  • 22 tools in one readout, Claude Code, Codex, Cursor, Copilot and Gemini CLI among them.
  • Several Claude and Codex accounts side by side on Pro, on Mac and Windows.
  • iPhone and Apple Watch companion, in TestFlight beta: widgets, the Dynamic Island and the Watch, synced from your Mac.
  • Codex banked resets listed with their expiry, with an alert before one expires and, on the Mac, applied after you confirm.
  • Support from the developer by email, and a 14-day refund on every purchase.

What it does not do: run on Linux, export data, show the session context window, or replace your provider's billing page. The 7-day trial unlocks all of Pro with no card, so the cheapest test is to install it next to whatever you use now.

Review SessionWatcher for Claude
SessionWatcher

Claude Code and 21 more tools.
Mac menu bar or Windows taskbar.

Native app for the Mac menu bar and the Windows taskbar. Track Claude, Codex and 20 more tools' usage, costs, and rate limits in real time.

★★★★★Trusted by developers daily
nicojerome

“Fast, simple, and does exactly what it should. Definitely worth it.”

@nicojerome on GitHub

Download Free for Mac

macOS 14+. Free for 7 days, no credit card.

Claude Code usage trackers on Windows

Most of this guide is not Mac-specific. /usage and the custom status line work wherever Claude Code runs, the Claude Console works in any browser, and ccusage and CCUM are terminal tools. For a reading outside the terminal on Windows, the options checked October 6, 2026:

  • Claude Code Usage Monitor by CodeZeno: free, MIT, a taskbar widget for Claude Code, Codex, Antigravity, OpenCode Go, Cursor, Grok Build and Copilot, with several Claude Code and Codex accounts; install with winget. Latest release v2.18.1, October 5, 2026.
  • Usage Monitor for Claude: free, MIT, a Claude-only tray app for Windows and Linux with threshold and reset notifications; a portable exe.
  • OpenQuota: free, MIT, a tray app for Windows, macOS and Linux covering twelve providers; its installers are not code-signed.
  • Win-CodexBar: free, MIT, a community port of CodexBar to the Windows tray, more than 50 providers per its README.
  • SessionWatcher for Windows (ours, paid): the readout inside the taskbar on Windows 11, or a small window beside it on Windows 10 (version 2004 or later), 22 tools, notifications at the thresholds you set, and one licence for your Mac too.

The full field, with install notes and caveats, is in our guide to AI usage trackers for Windows.

Frequently asked questions

What is the best Claude usage tracker for Mac?

It depends on what else you track. For Claude alone and free, Claude Usage Tracker is an open-source menu bar app with unlimited Claude accounts, per-model limits and Codex support. For many providers and free, CodexBar lists more than 100 in its README and also runs on Linux. Usagebar is a $9 Claude-only app that also warns when a session’s context runs low. SessionWatcher (ours, paid) covers 22 AI coding tools in the Mac menu bar and the Windows taskbar on one licence, with several Claude and Codex accounts and an iPhone and Apple Watch companion in beta. Checked against each project’s own README or site on November 15, 2026.

Does Claude Code already show plan-limit information?

Yes, in two documented places. On a Pro, Max, Team, or Enterprise plan, Anthropic documents plan usage bars and a usage breakdown on the /usage screen, with a toggle between the last 24 hours and the last 7 days. A custom Claude Code status line can also receive five-hour and seven-day rate-limit percentages and reset timestamps; Anthropic documents those rate_limits fields as appearing only for Claude.ai Pro and Max subscribers (or behind a Claude apps gateway with a spend limit), after the first API response in a session, and each window can be independently absent. Checked against Anthropic documentation on October 6, 2026.

Is the cost shown by /usage my Anthropic bill?

No. Anthropic says Claude Code calculates the session dollar figure locally from token counts at list price. It can differ from promotional pricing, contracted discounts, or your actual bill. The Claude Console Usage page is authoritative for API billing; subscription users have usage included in their plan unless they turn on usage credits.

Can Claude Code show usage in a custom status line?

Yes. Anthropic documents a configurable status line, generated with the /statusline command or configured manually through the statusLine setting. Claude Code sends the script JSON session data documented to include cost.total_cost_usd, context-window fields, and rate_limits fields when they are available. It stays inside Claude Code, and you own the script that chooses and formats the fields.

Does ccusage still cover only Claude Code?

No. The ccusage README, checked on October 6, 2026 with v20.0.26 (September 27, 2026) as the latest release, lists 18 agent CLIs, including Claude Code, Codex, OpenCode, GitHub Copilot CLI, Gemini CLI, Antigravity, Kimi, Qwen and Grok Build CLI; Cursor is not in the list. It also documents a Claude five-hour block report with active monitoring and a beta Claude Code status-line integration.

Is there a Claude Code usage tracker for Windows?

Yes, several. /usage and the status line work wherever Claude Code runs. Free Windows apps include CodeZeno’s Claude Code Usage Monitor (a taskbar widget that also covers Codex, Cursor and Copilot), Usage Monitor for Claude (a Claude-only tray app with threshold notifications) and OpenQuota (a tray app for Windows, macOS and Linux). SessionWatcher for Windows (ours, paid) puts Claude Code and 21 other tools inside the Windows 11 taskbar, or beside it on Windows 10 version 2004 or later, and one licence covers a Mac too.

Compared head to head

One page per tool, each starting with when that tool is the better choice than SessionWatcher.