LLM Token Tracker 🧮

Token usage tracker for OpenAI, Claude, and Gemini APIs with MCP (Model Context Protocol) support. Pass accurate API costs to your users.

![npm version](https://www.npmjs.com/package/llm-token-tracker) ![License: MIT](https://opensource.org/licenses/MIT)

✨ Features

🎯 Simple Integration - One line to wrap your API client
📊 Automatic Tracking - No manual token counting
💰 Accurate Pricing - Up-to-date pricing for all models (2025)
🔄 Multiple Providers - OpenAI, Claude, and Gemini support
📈 User Management - Track usage per user/session
🌐 Currency Support - USD and KRW
🤖 MCP Server - Use directly in Claude Desktop!
🆕 Intuitive Session Tracking - Real-time usage with progress bars

📦 Installation

npm install llm-token-tracker

🚀 Quick Start

Option 1: Use as Library

const { TokenTracker } = require('llm-token-tracker');
// or import { TokenTracker } from 'llm-token-tracker';

// Initialize tracker
const tracker = new TokenTracker({
  currency: 'USD' // or 'KRW'
});

// Example: Manual tracking
const trackingId = tracker.startTracking('user-123');

// ... your API call here ...

tracker.endTracking(trackingId, {
  provider: 'openai', // or 'anthropic' or 'gemini'
  model: 'gpt-3.5-turbo',
  inputTokens: 100,
  outputTokens: 50,
  totalTokens: 150
});

// Get user's usage
const usage = tracker.getUserUsage('user-123');
console.log(`Total cost: $${usage.totalCost}`);

🔧 With Real APIs

To use with actual OpenAI/Anthropic APIs:

const OpenAI = require('openai');
const { TokenTracker } = require('llm-token-tracker');

const tracker = new TokenTracker();
const openai = tracker.wrap(new OpenAI({
  apiKey: process.env.OPENAI_API_KEY
}));

// Use normally - tracking happens automatically
const response = await openai.chat.completions.create({
  model: "gpt-3.5-turbo",
  messages: [{ role: "user", content: "Hello!" }]
});

console.log(response._tokenUsage);
// { tokens: 125, cost: 0.0002, model: "gpt-3.5-turbo" }

Option 2: Use as MCP Server

Add to Claude Desktop settings (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "token-tracker": {
      "command": "npx",
      "args": ["llm-token-tracker"]
    }
  }
}

Then in Claude:

"Calculate current session usage" - See current session usage with intuitive format
"Calculate current conversation cost" - Get cost breakdown with input/output tokens
"Track my API usage"
"Compare costs between GPT-4 and Claude"
"Show my total spending today"

Available MCP Tools

get_current_session - 🆕 Get current session usage (RECOMMENDED)

Returns: Used/Remaining tokens, Input/Output breakdown, Cost, Progress bar
Default user_id: current-session
Default budget: 190,000 tokens
Perfect for real-time conversation tracking!

track_usage - Track token usage for an AI API call

Parameters: provider, model, input_tokens, output_tokens, user_id

get_usage - Get usage summary for specific user or all users

compare_costs - Compare costs between different models

clear_usage - Clear usage data for a user

Example MCP Output

💰 Current Session
━━━━━━━━━━━━━━━━━━━━━━
📊 Used: 62,830 tokens (33.1%)
✨ Remaining: 127,170 tokens
[██████░░░░░░░░░░░░░░]

📥 Input: 55,000 tokens
📤 Output: 7,830 tokens
💵 Cost: $0.2825
━━━━━━━━━━━━━━━━━━━━━━

📋 Model Breakdown:
  • anthropic/claude-sonnet-4.5: 62,830 tokens ($0.2825)

📊 Supported Models & Pricing (Updated 2025)

OpenAI (2025)

| Model | Input (per 1K tokens) | Output (per 1K tokens) | Notes | |-------|----------------------|------------------------|-------| | GPT-5 Series | | | | | GPT-5 | $0.00125 | $0.010 | Latest flagship model | | GPT-5 Mini | $0.00025 | $0.0010 | Compact version | | GPT-4.1 Series | | | | | GPT-4.1 | $0.0020 | $0.008 | Advanced reasoning | | GPT-4.1 Mini | $0.00015 | $0.0006 | Cost-effective | | GPT-4o Series | | | | | GPT-4o | $0.0025 | $0.010 | Multimodal | | GPT-4o Mini | $0.00015 | $0.0006 | Fast & cheap | | o1 Reasoning Series | | | | | o1 | $0.015 | $0.060 | Advanced reasoning | | o1 Mini | $0.0011 | $0.0044 | Efficient reasoning | | o1 Pro | $0.015 | $0.060 | Pro reasoning | | Legacy Models | | | | | GPT-4 Turbo | $0.01 | $0.03 | | | GPT-4 | $0.03 | $0.06 | | | GPT-3.5 Turbo | $0.0005 | $0.0015 | Most affordable | | Media Models | | | | | DALL-E 3 | $0.040 per image | - | Image generation | | Whisper | $0.006 per minute | - | Speech-to-text |

Anthropic (2025)

| Model | Input (per 1K tokens) | Output (per 1K tokens) | Notes | |-------|----------------------|------------------------|-------| | Claude 4 Series | | | | | Claude Opus 4.1 | $0.015 | $0.075 | Most powerful | | Claude Opus 4 | $0.015 | $0.075 | Flagship model | | Claude Sonnet 4.5 | $0.003 | $0.015 | Best for coding | | Claude Sonnet 4 | $0.003 | $0.015 | Balanced | | Claude 3 Series | | | | | Claude 3.5 Sonnet | $0.003 | $0.015 | | | Claude 3.5 Haiku | $0.00025 | $0.00125 | Fastest | | Claude 3 Opus | $0.015 | $0.075 | | | Claude 3 Sonnet | $0.003 | $0.015 | | | Claude 3 Haiku | $0.00025 | $0.00125 | Most affordable |

Google Gemini (2025)

| Model | Input (per 1K tokens) | Output (per 1K tokens) | Notes | |-------|----------------------|------------------------|-------| | Gemini 2.0 Series | | | | | Gemini 2.0 Flash (Exp) | Free | Free | Experimental preview | | Gemini 2.0 Flash Thinking | Free | Free | Reasoning preview | | Gemini 1.5 Series | | | | | Gemini 1.5 Pro | $0.00125 | $0.005 | Most capable | | Gemini 1.5 Flash | $0.000075 | $0.0003 | Fast & efficient | | Gemini 1.5 Flash-8B | $0.0000375 | $0.00015 | Ultra-fast | | Gemini 1.0 Series | | | | | Gemini 1.0 Pro | $0.0005 | $0.0015 | Legacy model | | Gemini 1.0 Pro Vision | $0.00025 | $0.0005 | Multimodal | | Gemini Ultra | $0.002 | $0.006 | Premium tier |

Note: Prices shown are per 1,000 tokens. Batch API offers 50% discount. Prompt caching can reduce costs by up to 90%.

🎯 Examples

Run the example: ``bash npm run example ``

Check examples/basic-usage.js for detailed usage patterns.

📝 API Reference

`new TokenTracker(config)`

config.currency: 'USD' or 'KRW' (default: 'USD')
config.webhookUrl: Optional webhook for usage notifications

`tracker.wrap(client)`

Wrap an OpenAI or Anthropic client for automatic tracking.

`tracker.forUser(userId)`

Create a user-specific tracker instance.

`tracker.startTracking(userId?, sessionId?)`

Start manual tracking session. Returns tracking ID.

`tracker.endTracking(trackingId, usage)`

End tracking and record usage.

`tracker.getUserUsage(userId)`

Get total usage for a user.

`tracker.getAllUsersUsage()`

Get usage summary for all users.

🛠 Development

# Install dependencies
npm install

# Build TypeScript
npm run build

# Watch mode
npm run dev

# Run examples
npm run example

📄 License

MIT

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

🐛 Issues

For bugs and feature requests, please create an issue.

📦 What's New in v2.4.0

🎉 Gemini API Support - Full integration with Google's Gemini models
💎 Gemini 2.0 Support - Free preview models included
📊 Enhanced Pricing - Up-to-date Gemini 1.5 and 2.0 pricing
🔧 Auto-detection - Automatic Gemini client wrapping
💰 Cost Comparison - Compare Gemini with OpenAI and Claude

📦 What's New in v2.3.0

💱 Real-time exchange rates - Automatic USD to KRW conversion
🌐 Uses exchangerate-api.com for accurate rates
💾 24-hour caching to minimize API calls
📊 New get_exchange_rate tool to check current rates
🔄 Background auto-updates with fallback support

What's New in v2.2.0

🗄️ File-based persistence - Session data survives server restarts
💾 Automatic saving to ~/.llm-token-tracker/sessions.json
🔄 Works for both npm and local installations
📊 Historical data tracking across sessions
🎯 Zero configuration required - just works!

What's New in v2.1.0

🆕 Added get_current_session tool for intuitive session tracking
📊 Real-time progress bars and visual indicators
💰 Enhanced cost breakdown with input/output token separation
🎨 Improved formatting with thousands separators
🔧 Better default user_id handling (current-session)

---

Built with ❤️ for developers who need transparent AI API billing.

llm-token-tracker