Why 86% of Claude Code Bill Has Nothing to Do With Your Prompts
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A detailed breakdown of where Claude Code token costs actually come from, based on a 30-day trace across a 45-person engineering team. Only 14% of input tokens are actual user prompts — the rest comes from static overhead, MCP server schemas, CLAUDE.md files, tool results, prior assistant context, and thinking tokens. The post explains each cost category, highlights the visibility gap in existing tools, and introduces Comet's Opik Cost Intelligence plugin as a solution. Practical audit steps are also provided: removing unused MCP servers, shrinking CLAUDE.md, switching to medium thinking effort, keeping sessions short, and defaulting to Sonnet over Opus for routine work.
