See exactly how you're using Claude Code — tokens, costs, models, and projects, all from your local data.
Claude Code writes detailed usage logs locally — token counts, models, sessions, projects — regardless of your plan. This dashboard reads those logs and turns them into charts and cost estimates. Works on API, Pro, and Max plans.
Originally created by: The Product Compass Newsletter
This fork (josepe98/claude-usage) is actively maintained and includes bug fixes and features not yet merged upstream. PRs welcome.
What's different here vs phuryn/claude-usage
- Fixed: Hourly chart crash — dashboard went blank when touching any filter (#74)
- Fixed: Blank dashboard for users with non-standard model names (#76, #106)
- Fixed: Bookmarkable URLs (
?range=30d) returning 404 (#80)- Fixed: Concurrent requests blocked — slow data loads no longer freeze the dashboard (#78)
- New: Appearance gallery with 5 bundled themes (Apple, Linear, Vercel, Notion, Stripe) at
/themes
Works on API, Pro, and Max plans — Claude Code writes local usage logs regardless of subscription type. This tool reads those logs and gives you visibility that Anthropic's UI doesn't provide.
Captures usage from:
- Claude Code CLI (
claudecommand in terminal) - VS Code extension (Claude Code sidebar)
- Dispatched Code sessions (sessions routed through Claude Code)
Also captured:
- Claude Desktop Cowork sessions — via local audit logs (see Cowork sessions below)
- Python 3.8+
- No third-party packages — uses only the standard library (
sqlite3,http.server,json,pathlib)
Anyone running Claude Code already has Python installed.
No pip install, no virtual environment, no build step.
git clone https://github.com/josepe98/claude-usage
cd claude-usage
python cli.py dashboard
git clone https://github.com/josepe98/claude-usage
cd claude-usage
python3 cli.py dashboard
macOS/Linux: use
python3. Windows: usepython. If neither works, see Requirements.
# Scan JSONL files and populate the database (~/.claude/usage.db)
python cli.py scan
# Show today's usage summary by model (in terminal)
python cli.py today
# Show the last 7 days (per-day breakdown + by-model totals)
python cli.py week
# Show all-time statistics (in terminal)
python cli.py stats
# Scan + open browser dashboard at http://localhost:8080
python cli.py dashboard
# Custom host and port via environment variables
HOST=0.0.0.0 PORT=9000 python cli.py dashboard
# Scan a custom projects directory
python cli.py scan --projects-dir /path/to/transcripts
# List available themes
python cli.py theme list
# Apply a theme (apple, linear, vercel, notion, stripe)
python cli.py theme set linear
The scanner is incremental — it tracks each file's path and modification time, so re-running scan is fast and only processes new or changed files.
By default, the scanner checks both ~/.claude/projects/ and the Xcode Claude integration directory (~/Library/Developer/Xcode/CodingAssistant/ClaudeAgentConfig/projects/), skipping any that don't exist. Use --projects-dir to scan a custom location instead.
Claude Code writes one JSONL file per session to ~/.claude/projects/. Each line is a JSON record; assistant-type records contain:
message.usage.input_tokens— raw prompt tokensmessage.usage.output_tokens— generated tokensmessage.usage.cache_creation_input_tokens— tokens written to prompt cachemessage.usage.cache_read_input_tokens— tokens served from prompt cachemessage.model— the model used (e.g.claude-sonnet-4-6)
scanner.py parses those files and stores the data in a SQLite database at ~/.claude/usage.db.
dashboard.py serves a single-page dashboard on localhost:8080 with Chart.js charts (loaded from CDN). It auto-refreshes every 30 seconds and supports model filtering with bookmarkable URLs. The bind address and port can be overridden with HOST and PORT environment variables (defaults: localhost, 8080).
Costs are calculated using Anthropic API pricing as of April 2026 (claude.com/pricing#api).
Only models whose name contains opus, sonnet, or haiku are included in cost calculations. Local models, unknown models, and any other model names are excluded (shown as n/a).
| Model | Input | Output | Cache Write | Cache Read |
|---|---|---|---|---|
| claude-opus-4-7 | $5.00/MTok | $25.00/MTok | $6.25/MTok | $0.50/MTok |
| claude-opus-4-6 | $5.00/MTok | $25.00/MTok | $6.25/MTok | $0.50/MTok |
| claude-sonnet-4-6 | $3.00/MTok | $15.00/MTok | $3.75/MTok | $0.30/MTok |
| claude-haiku-4-5 | $1.00/MTok | $5.00/MTok | $1.25/MTok | $0.10/MTok |
Note: These are API prices. If you use Claude Code via a Max or Pro subscription, your actual cost structure is different (subscription-based, not per-token).
Claude Desktop (the agent / Cowork mode in the desktop app) writes a per-session audit log to its userData directory:
| OS | Path |
|---|---|
| macOS | ~/Library/Application Support/Claude/local-agent-mode-sessions/ |
| Windows | %APPDATA%/Claude/local-agent-mode-sessions/ |
| Linux | $XDG_CONFIG_HOME/Claude/local-agent-mode-sessions/ (default ~/.config/...) |
scan automatically picks these up alongside ~/.claude/projects/. Sessions
appear in the dashboard under project names like Cowork/<8-char-id>.
Token totals come from the authoritative result.modelUsage blocks (the same
numbers Anthropic uses for billing), so cost estimates line up with what
the API reports rather than aggregating per-event streaming chunks.
| File | Purpose |
|---|---|
scanner.py |
Parses JSONL transcripts, writes to ~/.claude/usage.db |
dashboard.py |
HTTP server + single-page HTML/JS dashboard |
cli.py |
scan, today, stats, dashboard, theme commands |