How to check usage in Claude Code?
- Run
/usageinside a session. The Session block at the top shows token statistics for the current session; below it, subscribers get plan-limit bars. - Switch the window with
dfor the last 24 hours orwfor the last 7 days, and read the breakdown by skills, subagents, plugins and individual MCP servers. - Reconcile against the Usage page in the Claude Console if money matters — the in-session dollar figure is computed locally and isn't authoritative.
The number in your terminal is an estimate, and the docs say so.
Two caveats sit in Anthropic's own cost documentation, and both change how you should read the panel. First, Claude Code "computes the dollar figure locally from token counts priced at standard list rates, so it doesn't reflect promotional pricing or contracted discounts and may differ from your actual bill". For authoritative billing, the Console's Usage page is the reference.
Second, the plan-limit breakdown is "approximate and computed from local session history on this machine, so usage from other devices or claude.ai is not included". If you work across a laptop and a desktop, neither number is the whole picture.
What the breakdown does well is attribution: it splits recent usage across skills, subagents, plugins and individual MCP servers as percentages, and flags behaviours like long context or cache misses when one accounts for 10% or more. That's the part that actually tells you where the tokens went.
Scope of our check: this page is read off the documentation. /usage is a session command that draws an interactive panel, so it isn't scriptable and we have no transcript to publish — unlike the CLI-level commands we tested elsewhere.
Reading it correctly.
The Session block's cost figure is aimed at API users, who are billed per token. On Pro or Max your usage is included in the subscription, so that dollar number isn't a bill — it's a proxy for how hard you're driving the tool. What subscribers should read instead are the plan usage bars and the activity breakdown on the same screen.
If you're weighing which billing model suits you, the number is still useful as an input to the subscription-vs-API comparison.
Attribution by skills, subagents, plugins and MCP servers turns a vague "I hit my limit" into something actionable — a chatty MCP server or a subagent fan-out can quietly dominate. The flags for long context and cache misses point at the two biggest levers directly.
Context is the usual culprit: token cost scales with how much the model has to read, which is why prompt caching and compaction exist at all.
If the plan-limit request fails — usually because the usage endpoint is rate limited — /usage falls back to the last bars it loaded on that machine within the past hour, labelled Showing last-known usage with an age. Press r to retry. Without a recent snapshot it reports the rate limit and offers the same retry.
That's a display problem, not a billing one. The Console remains correct regardless.
Where the authoritative figures live.
The terminal panel is for steering during a session. Two places to look when the number has to be right.
Frequently asked.
Quick follow-ups people search after this question.
Q · 01 What's the command to see Claude Code usage? +
/usage, typed inside a running session. It shows a Session block with token statistics and, for subscribers, plan usage bars with a breakdown. Press d or w to switch between 24 hours and 7 days.