Claude Code Token Meter

claude-code-token-meter

Local-first dashboard for [Claude Code](https://claude.com/claude-code) token usage. Parses your local session logs (`~/.claude/projects/*/*.jsonl`) and serves an interactive viewer on `localhost`. **No API calls. Zero tokens spent on viewing.**


Why

Claude Code writes a JSONL log per session under `~/.claude/projects/`. Those files contain everything you need to understand your usage — input, output, cache-read and cache-create tokens per message — but there's no built-in way to explore them.

This tool parses them, aggregates per session, and exposes:

• Daily breakdown — weighted tokens per day
• Project breakdown — share of usage by project
• Model share — opus / sonnet / haiku split
• Top 30 sessions — biggest individual sessions, with the rest collapsed into a single "other sessions beyond top 30" line
• Date filter — 24h / 3d / 7d / 14d / 30d, switched without page reload
• Burn rate — rolling 5h / 24h / 7d weighted totals + 24h hourly sparkline
• Quota calibration — anchor the dashboard to your plan via the /usage % you see in Claude Code (see below); also auto-calibrates from clustered rate-limit hits stored locally
• Theme + lang — light / dark, EN / RU, persisted in localStorage

The "weighted" number uses Anthropic's published cost ratios as a proxy:

weighted = input × 1 + output × 5 + cache_create × 1.25 + cache_read × 0.1

It's not a literal subscription-quota percentage (Anthropic doesn't publish that formula), but it gives the right *relative* ranking of sessions and days.

Install

Requires Python 3.10+.

pipx install claude-code-token-meter

Or from source:

git clone https://github.com/itchernetski/claude-code-token-meter.git
cd claude-code-token-meter
python -m venv .venv
source .venv/bin/activate
pip install -e .

Run

claude-code-token-meter
claude-code-token-meter --port 8000
PORT=8000 claude