Visualizing Claude Code Token Usage
A new tool called 'codeburn' analyzes the session transcripts of Claude AI to provide visibility into how tokens are being used, including the breakdown of costs by task type, project, model, and more.
Why it matters
This tool gives Claude Code users much-needed visibility into their token usage, allowing them to identify areas for optimization and cost savings.
Key Points
- 1Codeburn reads the session transcripts stored by Claude Code to classify each conversation turn into 13 categories
- 2It shows the cost breakdown by task type (coding, debugging, exploration, brainstorming, etc.)
- 3Provides insights on token usage by project, model, tool, and MCP server
- 4Reveals that 56% of token spend is on 'conversation' with no tool usage, while actual coding is only 21%
Details
The author of this tool, who has been spending over $200 per day on Claude Code, wanted more visibility into where the tokens were being used. Codeburn analyzes the session transcripts stored by Claude Code and classifies each conversation turn into 13 different categories based on tool usage patterns. This allows the user to see a detailed breakdown of their token spend, including the costs by task type, project, model, tool, and MCP server. The tool also provides a daily activity chart with gradient bars to visualize the trends. One key insight was that 56% of the token spend was on 'conversation' turns where Claude was just responding with no tool usage, while the actual coding (edits, writes) was only 21% of the spend. This provides valuable information to optimize token usage and understand where the costs are coming from.
No comments yet
Be the first to comment