Claude Usage
Claude Usage is a local Python dashboard and CLI that scans Claude Code JSONL logs under ~/.claude/ and charts tokens, models, costs, and session history. We trial it when you want subscription-window visibility that Anthropic’s UI does not expose.
Blurb
Pro and Max subscribers get a progress bar. This gives you the full picture.
Summary
When to use: daily or weekly Claude Code spend reviews; comparing model mix across projects; Pro/Max quota tracking with progress bars; terminal summaries (today, week, stats) without opening a browser.
When to skip: deep per-turn debugging (use Claude DevTools); Cowork sessions (server-side, no local JSONL); multi-provider dashboards (CodexBar covers Claude plus Codex, Cursor, Gemini, and others from the menu bar).
Trade-offs: Python stdlib only (no pip deps), incremental SQLite scan, Homebrew/uv/Docker install paths, and an optional VS Code extension. Cost figures are estimates from local logs and published pricing, not exact invoices.
Details
| Topic | Notes |
|---|---|
| Upstream | phuryn/claude-usage (MIT, ~2k stars, Python) |
| Data source | Local JSONL under ~/.claude/ from Claude Code CLI, VS Code extension, and dispatched Code sessions |
| Not captured | Cowork sessions (server-side, no local transcripts) |
| Surfaces | Web dashboard (claude-usage dashboard), CLI subcommands, VS Code extension |
| Install | brew tap phuryn/claude-usage && brew install phuryn/claude-usage/claude-usage; or uv tool install git+https://github.com/phuryn/claude-usage; or clone + python3 cli.py dashboard |
| Database | Incremental scan into ~/.claude/usage.db (SQLite) |
| Companion | burnstop (same author ecosystem) |
CLI quick reference:
| |
Fit: Tool / AI Agent ops layer alongside Claude Code. Pairs with CodexBar when you need cross-provider menu-bar limits; pairs with Claude DevTools when you need session forensics, not aggregate charts.