Claude Code Usage Analytics

A local dashboard for your Claude Code history: tokens, activity, models, tools, agents and projects. Data is read directly from ~/.claude/projects/**/*.jsonl and never leaves your machine.

Unofficial community tool, not affiliated with or endorsed by Anthropic.

Features

• KPIs: total tokens, conversations, active sessions, agent calls, cache hit rate.
• Overview: tokens per day (input / output / cache write / cache read), activity heatmap, activity by hour.
• Models & tools: tokens per model, subagent calls (Agent), top tools.
• Projects: conversations and tokens per project, recent conversations.

Range: 7D / 30D / 90D / All. UI in English and Russian (toggle in the header).

Getting started

Requires Node.js ≥ 20.11.1.

npx claude-usage-analytics

or install globally:

npm install -g claude-usage-analytics
claude-usage-analytics      # http://127.0.0.1:3300, opens the browser

CLI options: -p, --port <port> (default 3300), --host <host> (default 127.0.0.1), --no-open, -h, --help.

The package ships only the prebuilt server, so installing it pulls no dependencies.

Development

git clone https://github.com/AlSovg/claude-usage-analytics.git
cd claude-usage-analytics
npm install
npm run dev        # http://localhost:3000
npm run build      # production build into .output
npm start          # run the build via the CLI

API

GET /api/analytics?range=7d|30d|90d|all (defaults to 30d; any other value → 400). A single endpoint returns all dashboard data; the response type is AnalyticsResponse in shared/types/analytics.ts.

How it's calculated

• Tokens — sum of message.usage from assistant messages. <synthetic> responses (interrupted streams) are excluded.
• Subagents — <session>/subagents/*.jsonl files count toward their parent session's tokens but not as separate conversations.
• User messages — real ones only: no isMeta, and origin.kind === "human" when origin is present.
• Dates and hours — in the machine's local time zone.
• Cache — each file's parse result is kept in memory keyed by mtime + size, so repeat requests only re-read changed files. The first request after server start is slower.

Limitations

• Claude Code deletes old transcripts itself (cleanupPeriodDays in ~/.claude/settings.json, 30 days by default), so the dashboard only sees what is still on disk.
• The claude.ai data export can't be used for stats: it has no token or model data.
• For "All", the heatmap is capped at the last 730 days (totals are unaffected).

Project structure

bin/cli.mjs                     CLI: starts the built server, opens the browser
server/api/analytics.get.ts     endpoint
server/utils/                   file scan, parser, cache, aggregation
shared/types/analytics.ts       types shared by client and server
app/pages/index.vue             dashboard page
app/components/analytics/       charts and widgets (Chart.js via vue-chartjs)
i18n/locales/                   ru / en translations