Claude Code Now Shows Exactly Where Your Usage Goes

In its June 12 update (2.1.174), Claude Code added a screen that breaks down exactly where your usage went. It shows consumption from cache misses, subagents, and even per-skill, per-plugin, and per-MCP usage. Right now it only appears in the VS Code extension.

01. Broken down from cache misses to per-skill usage

According to the changelog, usage attribution was added to the Account & usage screen (/usage) in VS Code. Over a 24-hour or 7-day window, it splits out how much was consumed by cache misses, long context, and subagents.

There's also a breakdown by skill, agent, plugin, and MCP. In other words, you can check item by item which skill or MCP server is eating the most tokens.

The old /usage screen was mostly about remaining quota. You couldn't tell why the quota was draining, and that gap has now been filled.

02. For now it only appears in the VS Code extension

The changelog marks it as a [VSCode] item only. There's no mention that the same breakdown landed in the terminal CLI's /usage.

To see it, open /usage in the VS Code extension. Your Claude Code version needs to be 2.1.174 or higher.

03. Useful right when quota management started to matter

Where your usage leaks is a real, practical problem for subscribers. In setups with lots of subagents or MCP tools, it was hard to gauge where consumption beyond your own typed messages was coming from.

The timing lines up, too. Per Anthropic's announcement, starting June 23, Fable 5 drops out of the free subscription tier and moves to separate usage credits. So a screen that shows what's burning credits was about to become necessary.

Exactly how fine-grained the tracking really gets is something you'll want to open it up and judge for yourself.

04. Model picker cleanup and admin settings came in too

In the /model picker, the issue where the model family that Default points to was hidden has been fixed. On Max, Team Premium, and Enterprise plans, Opus shows as a separate row; on Pro and Team plans, Sonnet does.

The 2.1.175 release that landed the same day added an enforceAvailableModels admin setting. When a team admin enforces an availableModels allowlist, the Default model is also restricted to the list, and the list can't be widened via personal or project settings.

Beyond that, smaller changes came along: a setting to disable mouse-wheel acceleration in fullscreen mode, and a fix for the 1-2 second hang on macOS and Linux when exiting right after interrupting a shell command.

Sources: Claude Code CHANGELOG (GitHub), Claude Code Docs — Changelog

This post summarizes official announcements and was not sponsored in any form by Anthropic.

---

Original with full infographics and visual structure: https://jessinvestment.com/claude-code-now-shows-exactly-where-your-usage-goes/