Home · Guides · Usage trackers compared

How to track your Claude Code usage: 7 tools compared

"Usage" means two different things — tokens you've spent (history) and quota you have left (live). Tools specialize in one or the other. Here's the whole landscape, including the built-in option, so you can pick by scenario instead of by GitHub stars.

Checked: July 13, 2026 · we build one of these tools, table stays honest anyway

First: what the built-in /usage shows

Claude Code ships its own answer — run /usage in a session and you get your 5-hour and weekly percentages with reset countdowns (in older builds: /status). claude.ai → Settings → Usage shows the same on the web. That's the official baseline, and for occasional checking it's fine. Everything below exists because of what it can't do: it lives inside a session, shows one account, keeps no history, and never warns you first.

The comparison

ToolFormatLive quota (5h + weekly)History / costAlertsMulti-accountCodex too
Built-in /usageCLI commandYes, current accountNoNoNoNo
ccusageCLI reportsNo (log-based)ExcellentNoPartialHistorical
Claude-Code-Usage-MonitorTerminal dashboardEstimated from logsGood, with forecastingIn-terminalNoNo
CodexBarmacOS menu barYes (59 providers)LimitedCountdownsNot a core featureYes
ClaudeUsageBarmacOS menu barPartialNoThresholdsNoNo
SessionWatchermacOS menu barPartialLimitedYesNoNo
AI Status BarmacOS menu barYesNo (by design)System, 80/90% + burn-rate forecastYes, first-classYes, live

Pick by scenario

AI Status Bar icon
Both providers, every account, zero commands AI Status Bar puts your Claude and Codex windows in the menu bar with reset times and alerts. Free & open source, notarized by Apple.
Download for macOS

FAQ

Why do log-based trackers disagree with /usage?

Local logs only see what happened on this machine in Claude Code. Your plan's budget also counts claude.ai chat and other devices, and Anthropic's metering weighs models and cache reads differently than raw token counts. Log-derived percentages are estimates; live tools read the same numbers /usage shows.

Do any of these tools consume my quota by polling?

No — usage numbers are metadata, not model requests. Checking them doesn't burn tokens or count against your windows.

What about Windows and Linux?

ccusage and Claude-Code-Usage-Monitor are cross-platform. The menu-bar options on this list, including AI Status Bar, are macOS-native.