Claude Code /usage Now Shows a Per-Category Breakdown of Your Rate Limit Usage
Claude CodeClaude Code's /usage command now surfaces a per-category breakdown of exactly what is consuming your rate limits — breaking down usage across skills, subagents, plugins, and individual MCP servers. This addresses a long-standing transparency gap where users hitting limits had no way to identify which components of their workflow were responsible. Developers with complex multi-server or multi-agent setups can now pinpoint which parts of their configuration drive the most compute and optimize accordingly.
Sources & Mentions
4 external resources covering this update
A More Transparent View Into Your Rate Limits
One of the most persistent frustrations for power users of Claude Code has been a lack of visibility into what actually drives their usage limits. A developer running several MCP servers, background subagents, and custom skills could hit a limit and have no precise way of knowing which component was responsible. Version 2.1.149 addresses this directly by upgrading the /usage command to show a per-category breakdown.
What the Breakdown Shows
The updated /usage screen now segments usage across the major cost drivers in a Claude Code session:
- Skills — how much of your limit budget is attributed to skill invocations
- Subagents — usage consumed by background or delegated subagent processes
- Plugins — contribution from installed plugin activity
- Per-MCP-server cost — an individual breakdown for each connected MCP server
This granularity is significant. MCP servers have been a documented source of token overhead: each connected server loads tool definitions into the context window, with a typical four-server setup costing roughly 7,000 tokens of overhead per turn before any productive work begins. With individual server visibility, developers can now identify which servers carry the heaviest footprint and make informed decisions about which connections are worth keeping active.
Why This Matters
Claude Code's rate limits are compute-based rather than message-based, which means a session running a complex agent pipeline with five MCP servers and several parallel subagents can exhaust limits much faster than a simple interactive session. Previously, debugging this required guesswork or external monitoring tools. The new per-category breakdown brings this intelligence natively into the CLI, accessible with a single command.
For teams using Claude Code at scale — with organization-wide plugins, shared MCP configurations, and distributed subagent workflows — this visibility is essential for budgeting and for spotting runaway processes before they drain the day's allocation.
Additional Improvements in 2.1.149
Version 2.1.149 also ships several smaller quality-of-life improvements. The /diff detail view is now keyboard-scrollable using arrow keys, j/k vi bindings, PgUp/PgDn, Space, Home, and End — matching the navigation conventions found elsewhere in the interface. Markdown output now renders GitHub Flavored Markdown (GFM) task list checkboxes, so - [ ] todo and - [x] done items display as visual checkboxes rather than plain bullet text.