🎯 Claude Code Usage Monitor

A privacy-first Claude Usage-Ops companion for Claude Code. It combines a Rich live terminal monitor with official statusline rate_limits, machine-readable state/export output, provenance labels, forecasting, and an opt-in local usage warehouse.

---

📑 Table of Contents

• ✨ Key Features
• 🚀 Installation
• ⚡ Modern Installation with uv (Recommended)
• 📦 Installation with pip
• 🛠️ Other Package Managers
• 📖 Usage
• Get Help
• Basic Usage
• Configuration Options
• Available Plans
• 🙏 Please Help Test This Release!
• ✨ Features & How It Works
• Current Features
• Understanding Claude Sessions
• Token Limits by Plan
• Smart Detection Features
• 🚀 Usage Examples
• Common Scenarios
• Best Practices
• 🔧 Development Installation
• Troubleshooting
• Installation Issues
• Runtime Issues
• 📞 Contact
• Related Tools
• 📚 Additional Documentation
• 📝 License
• 🤝 Contributors
• 🙏 Acknowledgments

✨ Key Features

🚀 v4.0.0 Major Update - Usage Ops Companion

• 🔎 Official-limit trust layer - --statusline captures Claude Code's official rate_limits; stale or expired captures fall back to labeled local estimates.
• 📦 Machine-readable protocol - --once, --compact, and --write-state all use one versioned snapshot builder with automation exit codes.
• 🏷️ Provenance labels - exported and displayed numbers distinguish official, local_estimate, experimental, and unknown confidence.
• 📈 Persistent usage warehouse - opt-in local history survives Claude's 30-day cleanup with project/model/day dimensions and CSV/JSON reports.
• 🧭 Forecasting and pace - reset-aware pace, date-context forecasts, official-only weekly percentages, and limit-hit freeze behavior.
• 🧩 Multi-source input - --data-paths, CLAUDE_CONFIG_DIR, and WSL discovery can scan multiple directories without merging unrelated accounts into one 5-hour window.
• 🖥️ Rich UI parity - live Rich output, rich one-shot output, compact output, state files, and exports use the same snapshot contract.
• 🧰 External companion boundary - GUIs, trays, provider adapters, and status bars should consume --write-state or --once --output json.
• 🧪 Regression coverage - the non-integration suite now covers the trust layer, state protocol, warehouse, reports, title updates, multi-source paths, and timezone edge cases.

📋 Default Custom Plan

The Custom plan is now the default option, specifically designed for 5-hour Claude Code sessions. It monitors three critical metrics: - Token usage - Tracks your token consumption - Messages usage - Monitors message count - Cost usage - The most important metric for long sessions

The Custom plan automatically adapts to your usage patterns by analyzing all your sessions from the last 192 hours (8 days) and calculating personalized limits based on your actual usage. This ensures accurate predictions and warnings tailored to your specific workflow.

🚀 Installation

⚡ Modern Installation with uv (Recommended)

Why uv is the best choice: - ✅ Creates isolated environments automatically (no system conflicts) - ✅ No Python version issues - ✅ No "externally-managed-environment" errors - ✅ Easy updates and uninstallation - ✅ Works on all platforms

The fastest and easiest way to install and use the monitor:

Install from PyPI

# Install directly from PyPI with uv (easiest)
uv tool install claude-monitor

# Run from anywhere
claude-monitor  # or cmonitor, ccmonitor for short

Install from Source

# Clone and install from source
git clone https://github.com/Maciek-roboblog/Claude-Code-Usage-Monitor.git
cd Claude-Code-Usage-Monitor
uv tool install .

# Run from anywhere
claude-monitor

First-time uv users

If you don't have uv installed yet, get it with one command:

# On Linux/macOS:
curl -LsSf https://astral.sh/uv/install.sh | sh

# On Windows:
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# After installation, restart your terminal

📦 Installation with pip

# Install from PyPI
pip install claude-monitor

# If claude-monitor command is not found, add ~/.local/bin to PATH:
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc  # or restart your terminal

# Run from anywhere
claude-monitor  # or cmonitor, ccmonitor for short

⚠️ PATH Setup: If you see WARNING: The script claude-monitor is installed in '/home/username/.local/bin' which is not on PATH, follow the export PATH command above.

⚠️ Important: On modern Linux distributions (Ubuntu 23.04+, Debian 12+, Fedora 38+), you may encounter an "externally-managed-environment" error. Instead of using --break-system-packages, we strongly recommend: 1. Use uv instead (see above) - it's safer and easier 2. Use a virtual environment - python3 -m venv myenv && source myenv/bin/activate 3. Use pipx - pipx install claude-monitor

See the Troubleshooting section for detailed solutions.

🛠️ Other Package Managers

pipx (Isolated Environments)

# Install with pipx
pipx install claude-monitor

# Run from anywhere
claude-monitor  # or claude-code-monitor, cmonitor, ccmonitor, ccm for short

conda/mamba

# Install with pip in conda environment
pip install claude-monitor

# Run from anywhere
claude-monitor  # or cmonitor, ccmonitor for short

📖 Usage

Get Help

# Show help information
claude-monitor --help

Available Command-Line Parameters

Parameter	Type	Default	Description
--plan	string	custom	Plan type: pro, max5, max20, team, or custom
--custom-limit-tokens	int	None	Token limit for custom plan (must be > 0)
--view	string	realtime	View type: realtime, daily, monthly, session, entries, sessions, or burn-rate
--output	string	rich	Output format: rich, json, text, or csv
--once	flag	False	Measure once, print a snapshot, and exit
--compact	flag	False	Single-line compact output for status bars
--write-state	flag	False	Write the snapshot to a state file for external tools
--state-file	path	None	State file path for --write-state
--statusline	flag	False	Run as a Claude Code statusline hook and capture official rate_limits
--api	flag	False	Enable the opt-in experimental Anthropic OAuth usage API
--data-paths	list	[]	Claude data directories to scan; repeat or comma-separate values
--warehouse	flag	False	Persist usage entries to the opt-in local warehouse
--warehouse-file	path	None	Usage warehouse file path
--warehouse-retention-days	int	365	Days of warehouse records to retain
--timezone	string	auto	Timezone (auto-detected). Examples: UTC, America/New_York, Europe/London
--time-format	string	auto	Time format: 12h, 24h, or auto
--theme	string	auto	Display theme: light, dark, classic, or auto
--refresh-rate	int	10	Data refresh rate in seconds (1-60)
--refresh-per-second	float	0.75	Display refresh rate in Hz (0.1-20.0)
--reset-hour	int	None	Daily reset hour (0-23)
--date-format	string	None	Date format for daily/monthly table periods
--abbreviate-tokens	flag	False	Abbreviate token counts in table views
--sparklines	flag	False	Show opt-in sparklines in table views
--filter-models	string	all	Use all models, or anthropic to exclude routed non-Claude models
--set-terminal-title	flag	False	Set the terminal title from the usage snapshot
--title-format	string	"{pct}% {plan}"	Terminal title template using pct, plan, used, limit, cost, reset
--hide-model-distribution	flag	False	Hide the model distribution bar
--no-header	flag	False	Hide the header banner
--no-emoji	flag	False	Render plain output without emoji
--log-level	string	INFO	Logging level: DEBUG, INFO, WARNING, ERROR, CRITICAL
--log-file	path	None	Log file path
--debug	flag	False	Enable debug logging
--version, -v	flag	False	Show version information
--clear	flag	False	Clear saved configuration

Plan Options

Plan	Token Limit	Cost Limit	Description
pro	19,000	$18.00	Claude Pro subscription
max5	88,000	$35.00	Claude Max5 subscription
max20	220,000	$140.00	Claude Max20 subscription
team	estimate	unverified	Team label; prefer official statusline or --plan custom
custom	P90-based	(default) $50.00	Auto-detection with ML analysis

Command Aliases

The tool can be invoked using any of these commands: - claude-monitor (primary) - claude-code-monitor (full name) - cmonitor (short) - ccmonitor (short alternative) - ccm (shortest)

Save Flags Feature

The monitor automatically saves your preferences to avoid re-specifying them on each run:

What Gets Saved: - View type (--view) - Plan (--plan) - Theme preferences (--theme) - Timezone settings (--timezone) - Time format (--time-format) - Refresh rates (--refresh-rate, --refresh-per-second) - Reset hour (--reset-hour) - Custom token limits (--custom-limit-tokens)

Configuration Location: ~/.claude-monitor/last_used.json

Usage Examples:

# First run - specify preferences
claude-monitor --plan pro --theme dark --timezone "America/New_York"

# Subsequent runs - preferences automatically restored
claude-monitor --plan pro

# Override saved settings for this session
claude-monitor --plan pro --theme light

# Clear all saved preferences
claude-monitor --clear

Key Features: - ✅ Automatic parameter persistence between sessions - ✅ CLI arguments always override saved settings - ✅ Atomic file operations prevent corruption - ✅ Graceful fallback if config files are damaged - ✅ Plan is saved and restored between sessions (override any time with --plan) - ✅ An explicitly chosen theme is kept and not overwritten by auto-detection

Basic Usage

With uv tool installation (Recommended)

# Default (Custom plan with auto-detection)
claude-monitor

# Alternative commands
claude-code-monitor  # Full descriptive name
cmonitor             # Short alias
ccmonitor            # Short alternative
ccm                  # Shortest alias

# Exit the monitor
# Press Ctrl+C to gracefully exit

Development mode

If running from source, use python -m claude_monitor from the src/ directory.

Machine-readable Usage Protocol

Use these surfaces when another tool needs current usage without scraping the Rich TUI:

# One-shot JSON snapshot with source/confidence/provenance fields
claude-monitor --once --output json

# Compact one-line status output
claude-monitor --once --compact

# Keep an atomically-written state file up to date for status bars and dashboards
claude-monitor --write-state --state-file ~/.claude-monitor/state/latest.json

# Install as a Claude Code statusline hook to capture official rate_limits
claude-monitor --statusline

--once exits with automation-friendly codes: 0 ok, 10 near limit, 11 limit hit, 20 indeterminate/no active session, and 30 no data or config error. When official statusline data is fresh it wins; otherwise the snapshot falls back to a labeled local estimate.

Persistent Usage Warehouse

The warehouse is opt-in and local-only. It stores versioned records by source, account, project, model, and day so history can outlive Claude's 30-day cleanup.

# Start persisting usage locally
claude-monitor --warehouse

# Export warehouse-backed entries as JSON
claude-monitor --warehouse --view entries --output json

# Export session percentiles and burn-rate reports as CSV
claude-monitor --warehouse --view sessions --output csv
claude-monitor --warehouse --view burn-rate --output csv

Configuration Options

Specify Your Plan

```bash

Custom plan with P90 auto-detection (Default)

claude-monitor --plan custom

Pro plan (~44,000 tokens)

claude-monitor --plan pro

Max5 plan (~88,000 tokens)

claude-monitor --plan max5

Max20 plan (~220,000 tokens)

claude-monitor --plan m