How to check Claude Code usage: five ways, fastest first

The five ways

Inside the CLI

/usage

d   last 24 hours
w   last 7 days

/context

Put quota in your status line

This is the highest-value option and almost nobody uses it. Claude Code runs a shell command to render its status line and pipes a JSON blob of session state to it on stdin. That blob includes your live plan percentages.

{
  "rate_limits": {
    "five_hour":  { "used_percentage": 23.5, "resets_at": 1738425600 },
    "seven_day":  { "used_percentage": 41.2, "resets_at": 1738857600 }
  },
  "cost":   { "total_cost_usd": 0.01234 },
  "model":  { "display_name": "Opus" },
  "context_window": { "used_percentage": 8 }
}

#!/usr/bin/env bash
input=$(cat)

model=$(jq -r '.model.display_name'                      <<< "$input")
dir=$(  jq -r '.workspace.current_dir'                   <<< "$input" | xargs basename)
h5=$(   jq -r '.rate_limits.five_hour.used_percentage // 0' <<< "$input" | cut -d. -f1)
d7=$(   jq -r '.rate_limits.seven_day.used_percentage // 0' <<< "$input" | cut -d. -f1)
cost=$( jq -r '.cost.total_cost_usd // 0'                <<< "$input")

printf "%s | %s | 5h %s%% | 7d %s%% | $%.2f" \
  "$model" "$dir" "$h5" "$d7" "$cost"

{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh",
    "padding": 2
  }
}

/statusline show model, 5 hour and weekly limit percentages, and a context bar

Full history from the session files

Claude Code writes a JSONL transcript for every session. ccusage reads them and aggregates, and needs no installation:

# daily totals across every detected agent CLI
npx ccusage@latest

# other groupings
npx ccusage@latest weekly
npx ccusage@latest monthly
npx ccusage@latest session

# just Claude Code, grouped by project
npx ccusage@latest claude daily --instances

# the current 5-hour block, with projections
npx ccusage@latest blocks --active

This works retroactively, because the data was already being written before you installed anything. The first time anyone runs it they usually discover weeks of history they did not know they had.

When two answers disagree

You will eventually get two different numbers for the same week and assume one tool is broken. Usually neither is. These are the five reasons, in the order they actually occur.

Quota is different. A status line re-runs only on events, such as a new assistant message or a refreshInterval tick, so between turns it is displaying the last values it was handed while a fresh /usage call asks now. A small gap between the two is normal and closes on your next message.

For a team or an organisation

None of the above scales past one machine, because the transcripts are local. Which route you take depends on how your organisation signs in.

curl "https://api.anthropic.com/v1/organizations/usage_report/claude_code?starting_at=2026-08-01&limit=100" \
  -H "anthropic-version: 2023-06-01" \
  -H "x-api-key: $ADMIN_API_KEY"

One call returns one UTC day of per-user records: sessions, lines added and removed, commits, pull requests, tool acceptance rates, and token and cost figures per model. Data lands with up to an hour of delay, and the endpoint is free.

Questions people ask

Keep reading