Claudetop: Real-Time Cost Monitoring for Claude Code Sessions

Claudetop is a monitoring tool for Claude Code that displays real-time cost tracking, cache efficiency metrics, and model comparisons during coding sessions. The tool was created after the developer noticed a discrepancy where model estimates showed $10 but actual billing was $65, with compaction hiding 80% of token usage.

Installation Options

Three installation methods are available:

Clone and install: git clone https://github.com/liorwn/claudetop.git then cd claudetop && ./install.sh
One-liner: curl -fsSL https://raw.githubusercontent.com/liorwn/claudetop/main/install.sh | bash
As a Claude Code plugin: claude plugin marketplace add liorwn/claudetop then claude plugin install claudetop

The plugin installation provides SessionEnd hook plus slash commands automatically: /claudetop:stats, /claudetop:dashboard, /claudetop:branch, /claudetop:export, and /claudetop:pricing.

Display Information

During sessions, Claudetop shows:

Current project location and depth
Active model and session duration
Current cost, hourly burn rate, and projected monthly cost
Cache efficiency percentage and cost per line
Comparative costs for other models (Opus, Sonnet, Haiku)
Token breakdown: input/output percentages with fresh, cwrite, and cread categories

Key Features

Real-time cost tracking: Shows actual session cost (green), burn rate per hour, and monthly forecast extrapolated from history. Pricing updates automatically from pricing.json when Anthropic changes prices.

Model cost comparison: Displays what the session would cost on Opus, Sonnet, or Haiku with cache-aware pricing that accounts for actual cache hit ratio. The current model is bolded for instant comparison.

Cache efficiency monitoring: Cache hit ratio indicates efficiency: green (≥60%) means most input tokens are being reused, red (<30%) indicates something is forcing full re-reads.

Smart Alerts

Alerts appear only when attention is needed:

$5 MARK / $10 / $25: Cost milestone crossed - gut-check value
OVER BUDGET: Daily budget exceeded - wrap up or switch models
CONSIDER FRESH SESSION: >2hrs + >60% context - diminishing returns
LOW CACHE: <20% cache after 5min - context was reset, tokens being re-read
BURN RATE: >$15/hr velocity - runaway subagents or tight loops
SPINNING?: >$1 spent, zero code output - stuck in research loop
TRY /fast: >$0.05/line on Opus - task doesn't need biggest model
COMPACT SOON: Context window >80% full - auto-compaction imminent

Session History & Analytics

Every session is automatically logged. Use claudetop-stats with various arguments:

claudetop-stats - Today's summary
claudetop-stats week - This week
claudetop-stats month - This month
claudetop-stats all - All time
claudetop-stats tag auth - Filter by tag

This tool is for developers using Claude Code who want visibility into their AI spending patterns and want to optimize model usage and cache efficiency.

📖 Read the full source: HN AI Agents