I measured what my Claude Code setup costs on every request. Most of my skills had gone blank.

7.9 relevance

Directly relevant to AI coding agent costs, offering actionable data for optimization.

AI/ML dev.to

Answer the board on AI readiness

Arrive with evidence, not opinion. See how your foundation benchmarks against your peers, get a clear read, and leave with a ranked place to start.

Why It Matters

For platform and AI engineers, this quantifies how accumulated tooling overhead degrades agent performance and inflates costs, directly impacting the viability of AI-assisted development workflows.

Key Takeaways

Audit your Claude Code setup with the 'deadweight' CLI and prune unused plugins, MCP servers, and skills to reclaim context window and reduce per-request token costs.

I measured what my Claude Code setup costs on every request. Most of my skills had gone blank.

Summary

Claude Code adds ~22k tokens of overhead per request from plugins, MCP servers, and skills — 74,500 tokens total on the first call — with skill descriptions truncated beyond ~30k characters, leaving 220 of 283 skills as bare names. The author's open-source CLI 'deadweight' audits these transcripts to identify deadweight like unused MCP servers and duplicate skills, enabling users to reduce context bloat and improve model performance.

Author

Murat Sert