使用说明
MCP Cost Tracker
An open-source MCP server that automatically tracks LLM API call costs, token usage, and generates visual dashboards.
mcp-cost-tracker is a powerful, easy-to-use MCP server designed for developers who want to monitor and control their LLM API spending. It provides a comprehensive suite of tools to track token usage, calculate costs across 50+ models, set budgets, and visualize spending patterns.
(This is a placeholder image. A real screenshot will be added after the first release.)
✨ Key Features
- Comprehensive Cost Tracking: Automatically calculates costs for 50+ models from major providers including OpenAI, Anthropic, Google, DeepSeek, Mistral, and more.
- 15+ MCP Tools: A rich set of tools for tracking usage, analyzing costs, setting budgets, comparing models, and exporting data.
- Visual Dashboard: Generate a beautiful, interactive HTML dashboard with charts for daily trends, cost breakdowns by model/provider/project, and top expensive requests.
- Budget Management: Set daily, weekly, or monthly budgets with alerts to prevent cost overruns.
- SQLite Persistence: All usage data is stored locally in a SQLite database, ensuring data ownership and enabling historical analysis.
- Easy Integration: Use as a standalone MCP server with any compatible client (like Claude Desktop) or integrate into your existing applications.
- Non-Intrusive: Designed to work without modifying your existing API call logic.
🚀 Getting Started
Installation
npm install -g mcp-cost-tracker
Running the Server
Simply run the command in your terminal:
mcp-cost-tracker
This will start the MCP server, listening on stdio for requests from a compatible client.
Connecting with Claude Desktop
- Open Claude Desktop settings.
- Go to the "MCP Servers" section.
- Click "Add Server" and choose "From Command".
- Enter
mcp-cost-trackeras the command. - Save and enable the server.
Now you can use all the cost tracking tools directly within Claude!
🛠️ MCP Tools
mcp-cost-tracker provides a rich set of tools to manage your LLM costs. Here are some of the key tools available:
| Tool Name | Description |
|---|---|
track_usage |
Record an API call with token counts and automatically calculate its cost. |
get_summary |
Get a summary of total costs, tokens, and request counts, with filters. |
get_cost_by_model |
See a cost breakdown grouped by model to identify top cost drivers. |
get_daily_trend |
View daily cost trends to spot spending patterns. |
set_budget |
Create or update a spending budget with alerts. |
check_budget |
Check the current status of your budgets. |
generate_dashboard |
Generate an interactive HTML dashboard. |
lookup_pricing |
Look up pricing for any supported model. |
estimate_cost |
Estimate the cost of an API call before making it. |
compare_models |
Compare the cost of a workload across different models. |
export_data |
Export usage data to CSV or JSON for external analysis. |
For a full list of tools and their arguments, you can use the tools/list method from your MCP client.
📊 Dashboard
To generate the HTML dashboard, use the generate_dashboard tool:
> generate_dashboard
This will create a dashboard.html file in your ~/.mcp-cost-tracker/ directory. Open this file in your browser to see a detailed visualization of your spending.
⚙️ Configuration
By default, the database is stored at ~/.mcp-cost-tracker/costs.db.
You can specify a custom database path by setting the MCP_COST_TRACKER_DB environment variable:
export MCP_COST_TRACKER_DB=/path/to/your/costs.db
mcp-cost-tracker
🤝 Contributing
Contributions are welcome! Please feel free to submit a pull request or open an issue to discuss improvements.
📄 License
This project is licensed under the MIT License. See the LICENSE file for details.
