🎯 claude-plan-usage

Your Claude plan usage limits — in the terminal. No more switching to the desktop app.

  Plan usage limits  Team

  Current session
  Resets in 4 hr 38 min
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━  13% used

  Weekly limits

  All models
  Resets in 130 hr 58 min
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━  4% used

  Last updated: 14:01:55  ● within limits

---

📑 Table of Contents

• ✨ Key Features
• 🤔 Why Not Other Tools?
• 🚀 Installation
• 📖 Usage
  • Terminal
  • Inside Claude Code (Slash Command)
  • Python API
• 🔍 How It Works
• 🎨 Progress Bar Colors
• 📋 Requirements
• ⚠️ Limitations
• 📝 License

---

✨ Key Features

• 🎯 Exact numbers — reads the same data as claude.ai/settings/usage, not estimates
• ⚡ Instant — one command, results in 2 seconds
• 🔌 Claude Code integration — add as a /usage slash command, never leave your session
• 🪶 Zero dependencies — just Python standard library, nothing to install
• 🔑 Zero config — auto-reads your auth from macOS Keychain
• 💰 Basically free — each check costs ~$0.000001 (one Haiku token)
• 📊 Color-coded progress bars — blue, yellow, red based on utilization

---

🤔 Why Not Other Tools?

Other usage trackers (like claude-monitor) parse your local session files and estimate your limits using token counts, ML predictions, and P90 calculations. They're guessing.

claude-plan-usage doesn't guess. It reads undocumented rate limit headers directly from Anthropic's API — the exact same data source that powers the claude.ai/settings/usage page.

When the dashboard says 50%, this tool says 50%. Same number. Same source.

	🔮 Other tools	🎯 claude-plan-usage
Data source	Local session files	✅ Live API response headers
Accuracy	Estimated from token counts	✅ Exact — same as claude.ai dashboard
What it shows	Token counts, burn rates, predictions	✅ Actual plan utilization % and reset times
Setup	Config files, plan selection, themes	✅ pip install and done
Dependencies	Heavy (ML, Rich, Pydantic, Sentry)	✅ Zero — Python standard library only
Complexity	100+ files	✅ ~200 lines of code

💡 The key discovery: Anthropic returns anthropic-ratelimit-unified-5h-utilization and anthropic-ratelimit-unified-7d-utilization headers on every API response. These are undocumented. Nobody was reading them. Now you can.

---

🚀 Installation

pip install claude-plan-usage

That's it. No config. No setup. No API keys needed.

---

📖 Usage

Terminal

claude-plan-usage

🔌 Inside Claude Code (Slash Command)

This is the cool part — you can check your usage without leaving your Claude Code session.

Create this file at ~/.claude/skills/usage/SKILL.md:

---
name: usage
description: Show Claude plan usage limits (current session and weekly)
disable-model-invocation: true
allowed-tools: Bash
---

Run the following command and display the output to the user exactly as-is:

\```bash
claude-plan-usage
\```

Now just type /usage in any Claude Code session. ✨

🐍 Python API

from claude_usage import get_plan_limits

limits, error = get_plan_limits()
if limits:
    print(f"Session: {limits['session']['percent']}% used")
    print(f"Weekly:  {limits['weekly']['percent']}% used")
    print(f"Status:  {limits['status']}")

---

🔍 How It Works

Anthropic returns undocumented rate limit headers on every API response:

Header	What it tells you
anthropic-ratelimit-unified-5h-utilization	📊 Current session usage (0.0 to 1.0)
anthropic-ratelimit-unified-5h-reset	⏰ When the session resets (unix timestamp)
anthropic-ratelimit-unified-7d-utilization	📊 Weekly usage (0.0 to 1.0)
anthropic-ratelimit-unified-7d-reset	⏰ When the week resets (unix timestamp)
anthropic-ratelimit-unified-status	✅ allowed or 🚫 throttled

How it works under the hood:

1. 🔑 Reads your OAuth token from macOS Keychain (stored by Claude Code CLI on login)
2. 📡 Sends one tiny Haiku API call (1 token)
3. 📋 Reads the rate limit headers from the response
4. 🎨 Renders progress bars with colors in your terminal

Zero config. Zero dependencies. Just works.

---

🎨 Progress Bar Colors

Color	Utilization	Vibe
🔵 Blue	0–49%	You're chilling
🟡 Yellow	50–79%	Maybe slow down
🔴 Red	80–100%	You're about to hit the wall

---

📋 Requirements

Requirement	Details
🍎 OS	macOS (uses Keychain for auth)
🤖 Claude Code	Installed and logged in
🐍 Python	3.9+

🐧 Linux/Windows users: PRs welcome! The main blocker is reading the OAuth token — everything else is cross-platform.

---

⚠️ Limitations

• macOS only for now — the OAuth token lives in macOS Keychain. Happy to accept PRs for other platforms.
• Undocumented API headers — Anthropic could change these anytime. If it breaks, open an issue.

---

📝 License

MIT

---

Built by @craakash — because checking usage shouldn't require opening another app.