Claude usage tracker
for Claude Code and claude.ai.

Where Claude shows your usage limits

Claude's paid plans meter usage in two windows: a session limit that resets every five hours, and a weekly limit. Anthropic counts claude.ai, Claude Code and Claude Desktop against the same limits, so a tracker that reads only one of them can show room you no longer have.

Anthropic shows your usage in three places:

• Settings > Usage, on the web and in Claude Desktop: a bar for the current session with the time left in it, bars for the weekly limits with their reset day, and any usage credits. On a usage-based Enterprise plan it shows consumption instead.
• /usage in Claude Code: the same plan bars on Pro, Max, Team and Enterprise, the session's tokens per model with a cost estimated at list price, and what used your limits over the last 24 hours or 7 days, by skill, subagent, plugin and MCP server. That breakdown comes from this machine's history, not from your other devices or claude.ai.
• The Usage page in the Claude Console, for API keys: Anthropic's docs call it the authoritative bill, where Claude Code's own cost figures are estimates.

Which Claude usage tracker to use

Pick a tracker by the question it answers: how much is left right now, where the tokens went, or how the whole team uses Claude.

A live meter for the five-hour limit

Claude Code hands every status line command a rate_limits object with the five-hour and weekly windows: the percent used, from 0 to 100, and when each one resets. It fills those two windows only for Pro and Max subscribers, after a session's first response, and behind a Claude apps gateway it carries your spend limit instead, so on a Team or Enterprise seat read /usage. The Claude Code status line guide has a recipe that prints the five-hour figure.

Two open-source tools build a fuller meter on the same numbers. Claude Status Bar draws five-hour and seven-day bars with reset countdowns and a projection to the end of each window, and adds a floating panel for the Claude desktop app on macOS. Claude Code Usage Monitor runs in a terminal of its own: it reads the official limits through the status line, falls back to local estimates it labels as estimates, and forecasts when you will hit the limit. Its default plan learns your limits from the last 8 days of sessions.

pip install claude-statusbar     # or: uv tool install / pipx install
cs --setup                       # wires the statusLine hook + installs the skill

uv tool install claude-monitor
claude-monitor  # or cmonitor, ccmonitor for short

Claude Code usage history and cost with ccusage

ccusage answers where the tokens went. It reads the session logs on your machine and prints tokens and an estimated cost by day, week, month or session, with a per-model breakdown, and its blocks report groups Claude Code's usage into 5-hour windows. The same tool reads Codex, Gemini CLI, GitHub Copilot CLI and 14 more agents, and --json hands the numbers to your own scripts.

Every local tracker has two blind spots. Its cost is a list-price estimate, while a subscriber pays for the seat (Claude Code pricing has what each plan includes). And it sees one machine: usage on claude.ai, on a second laptop or under a teammate's account isn't in its numbers. Claude Code also clears transcripts older than 30 days by default, which caps how far back a local report goes; Claude Code Usage Monitor's opt-in warehouse keeps its own history.

npx ccusage@latest                       # every agent it finds, by day
bunx ccusage claude monthly --breakdown  # Claude Code by month, per model

Tracking claude.ai chat usage

Chats on claude.ai draw on the same limits, but the Claude Code trackers above don't see them. Settings > Usage is the official view, and it is where Anthropic's occasional free limit reset appears: eligible plans sometimes get a Reset for free button that refills the five-hour or weekly limit, usable on the web or in Claude Desktop but not from Claude Code in a terminal.

The Claude Usage Tracker extension for Chrome and Firefox estimates the tokens behind each chat: uploaded files, project knowledge, your preferences, message history, system prompts and most connectors. It can't count web search results or Research, which run on Anthropic's side. To sync across devices it reads your organization ID from claude.ai and keeps usage in Firebase, so read its privacy policy before you install it on a work account.

Tracking Claude usage for a team

Team and Enterprise owners, and Console admins, get Claude Code analytics: daily active users and sessions, lines accepted and the accept rate, the top commands, and each member's lines accepted this month, exportable as a CSV. A Value tab turns those into an estimated productivity lift and a cost per commit, with every formula editable. The numbers refresh once a day and reset each month, and they include no per-engineer tokens or limit use; the API and OpenTelemetry routes to those are on the Claude Code usage dashboard guide.

Zest is the per-engineer view across tools. Each engineer installs the plugin (setup), and every Claude Code session lands on the team dashboard with its models, tokens and list-price cost, the five-hour and weekly limit use the status line reported, and the pull request it led to. Zest's status line passes through one you already run, such as Claude Status Bar's, and Codex and Cursor sessions sit on the same dashboard.

What Zest doesn't do: draw a live meter while you work, or count claude.ai chats, which use the same limits. Its limit figures come from Pro and Max accounts only, because those are the accounts Claude Code reports them for.