Claude Token Tracker

A local web dashboard for tracking your Claude Code token usage, costs, and cache efficiency. Zero dependencies — runs entirely on Python's standard library.

Dashboard showing accurate token tracking with recursive subagent scanning and resumed session deduplication fixes (September 2, 2026).

Latest Updates (September 2026)

Bug Fixes (Sept 2, 2026)

• Fixed recursive file scanning: Changed from glob("*.jsonl") to rglob("*.jsonl") to catch nested subagent transcripts in <session-id>/subagents/workflows/ directories (was missing 82% of actual token usage)
• Fixed resumed session double-counting: Implemented global requestId deduplication — when sessions are resumed with --resume, prior transcripts copy into new files with duplicate requestIds. Now takes max(input, output, cache_read, cache_write) across all copies instead of per-file deduplication
• Added Opus 5, Sonnet 5, Fable 5 pricing: Updated MODEL_PRICING for latest Claude models (Opus 5: $5/$25 input/output, Sonnet 5: $2/$10, Fable 5: $10/$50)

Impact: Token counter now accurately tracks 100% of usage (was only showing ~83% before due to missing subagent transcripts and resumed session bugs).

Features

• KPI Cards — Total tokens, estimated API cost, cache savings, and session count at a glance
• Time Range Filtering — View stats for Today, This Week, This Month, or All Time
• Daily Usage Chart — Stacked bar chart of input, output, cache write, and cache read tokens (toggle to cost view)
• Model Breakdown — Donut chart and table showing usage and cost per model (Opus, Sonnet, Haiku)
• Cache Efficiency — See how much caching saves you with hit rate and cost comparison
• Activity by Hour — Heatmap-style chart of when you use Claude most
• Project Breakdown — Token usage and cost grouped by project directory
• Session Drill-Down — Click any session to see per-request token and cost details
• Auto-Refresh — Optional 30-second auto-refresh to monitor usage in real time

How It Works

Claude Code stores session data as JSONL files in ~/.claude/projects/. The tracker:

1. Recursively scans all session files in ~/.claude/projects/ (including subagent transcripts in <session-id>/subagents/workflows/ directories)
2. Parses each JSONL file to extract token usage per request (input, output, cache read, cache write)
3. Deduplicates requestIds globally — resumed sessions create duplicate requestIds across transcript copies, tracker takes max usage to avoid double-counting
4. Computes estimated API costs using current Anthropic pricing (Opus 5, Sonnet 5, Fable 5, Haiku 4.5 tiers)
5. Serves a single-page dashboard on localhost:8050 with interactive Chart.js visualizations

All data stays local — nothing is sent anywhere.

Data Sources Tracked

• Main session transcripts: ~/.claude/projects/<project>/sessions/<session-id>.jsonl
• Subagent transcripts: ~/.claude/projects/<project>/sessions/<session-id>/subagents/workflows/wf_*/agent-*.jsonl (now included)
• Resumed sessions: Duplicates are merged using global requestId deduplication

Installation

# Clone the repo
git clone https://github.com/ColinM-sys/Claude-Token-Tracker.git
cd Claude-Token-Tracker

# Run it (Python 3.8+ required, no pip install needed)
python server.py

The dashboard opens automatically at http://localhost:8050.

Requirements

• Python 3.8+ (uses only the standard library)
• Claude Code installed (the dashboard reads from ~/.claude/)

No pip install, no requirements.txt, no Node.js — just Python.

Pricing Reference

Cost estimates use Anthropic's published API pricing for reference. Claude Code itself is a flat monthly subscription — the costs shown are what equivalent API usage would cost.

Current Models (2026)

Opus Tier (Most Capable)

Model	Input	Output	Cache Read	Cache Write (5min)
Claude Opus 5	$5.00/M	$25.00/M	$0.50/M	$6.25/M
Claude Opus 4.8	$5.00/M	$25.00/M	$0.50/M	$6.25/M
Claude Opus 4.6/4.5	$5.00/M	$25.00/M	$0.50/M	$6.25/M
Claude Opus 4.1/4	$15.00/M	$75.00/M	$1.50/M	$18.75/M

Sonnet Tier (Balanced)

Model	Input	Output	Cache Read	Cache Write (5min)
Claude Sonnet 5	$2.00/M	$10.00/M	$0.20/M	$2.50/M
Claude Sonnet 4.6/4.5/4	$3.00/M	$15.00/M	$0.30/M	$3.75/M

Fable Tier (Experimental)

Model	Input	Output	Cache Read	Cache Write (5min)
Claude Fable 5	$10.00/M	$50.00/M	$1.00/M	$12.50/M

Haiku Tier (Speed/Cost)

Model	Input	Output	Cache Read	Cache Write (5min)
Claude Haiku 4.5	$1.00/M	$5.00/M	$0.10/M	$1.25/M
Claude Haiku 3.5	$0.80/M	$4.00/M	$0.08/M	$1.00/M

Legend: M = per 1 million tokens. Bold = latest/newest model in each tier.

License

MIT

---

Author

Built by Colin McDonough — LinkedIn · GitHub