A new open-source TUI, Codeburn, analyzes Claude Code session transcripts to show token spend by task type, helping developers optimize their usage and costs.
The Problem: Blind Spending
Developers using Claude Code heavily, like the creator spending $200+ daily, have had zero visibility into what consumes their tokens. While tools like ccusage show aggregate cost, they don't answer critical questions: Is debugging the budget killer? Which project is the most expensive? Is most of the spend just on conversation?
The Solution: Codeburn
codeburn is a new, open-source Terminal User Interface (TUI) that solves this. It reads the session transcripts Claude Code already stores locally (~/.claude/projects/) and classifies every interaction into 13 deterministic categories based on tool usage patterns—no additional LLM calls required.
What It Shows You
- Cost by Task Type: Coding (edits, writes), Debugging, Exploration, Brainstorming, and—crucially—Conversation (turns with no tool use).
- Cost by Project, Model, Tool, and MCP Server: Pinpoint exactly which part of your workflow is driving costs.
- Daily Activity Chart: Visual timeline with gradient bars showing activity peaks.
- Interactive Views: Use arrow keys to switch between Today, Week, and Month summaries.
- Swiftbar Widget (macOS): A menu bar widget for at-a-glance daily spend.
The Eye-Opening Insight
For its creator, the data was revealing: 56% of total spend was on "Conversation"—turns where Claude responded without using any tools. The actual act of coding (file edits and writes) accounted for only 21%. This insight is a direct lever for optimizing prompts and workflow to reduce waste.
How To Use It Right Now
Installation and use is straightforward:
npx codeburn
That's it. It works with any existing Claude Code installation and requires no configuration. It parses your local session history on the fly.
Actionable Takeaways from the Data
- Audit Your "Conversation" Spend: High conversation costs often mean you're not giving Claude enough context or clear instructions upfront, leading to back-and-forth clarification. Be more specific in your initial prompt.
- Compare Project Costs: Identify if one legacy codebase or experimental project is disproportionately expensive. This can justify refactoring or changing your approach.
- Evaluate MCP Servers: See if a particular MCP server (like a database connector or API tool) is token-heavy. You might need to optimize its prompts or seek an alternative.
- Model Selection Validation: Confirm if using a more capable (and expensive) model like Claude Opus 4.6 is justified for a given task type, or if Sonnet would suffice.
This tool transforms Claude Code from a black-box expense into an instrumented, optimizable development environment.
Originally published on gentic.news
Top comments (0)