How to Check Claude Code Usage (2026): Every Method That Actually Works

The question sounds simple: how do you check Claude Code usage? In practice the answer is layered because Claude Code doesn't ship a single dashboard. It ships a couple of in-session commands, writes transcripts to disk, and leaves the aggregation to you.

This guide covers the four methods that actually work in 2026, in order of depth:

1. /status inside a session — shows the current 5-hour rolling window
2. /cost inside a session — shows the current conversation's token totals
3. Transcript files under ~/.claude/projects — the underlying per-turn data Claude Code writes to disk
4. Free Claude usage dashboard — rolls up the transcript files into daily, weekly, and monthly views with API-equivalent cost

The right method depends on what you're actually trying to answer. If it's "am I about to hit a cap?", the in-session commands are enough. If it's "should I be on Pro or Max?", you need the historical view.

Method 1: /status — the in-session snapshot

Inside a Claude Code session, type /status and hit enter. You get a snapshot of the current 5-hour window: how much of your subscription cap is remaining, broken out per model.

The output tells you the immediately actionable thing — whether you can keep pushing on Opus or need to route the rest of the work to Sonnet or Haiku. It does not tell you:

• What you consumed over the last day, week, or month
• Which project burned the most tokens
• Whether last week's usage would have justified an upgrade
• What the compute would have cost at API rates

/status is a live gauge, not a history. Treat it that way.

Method 2: /cost — the current session's token math

/cost, also typed inside a running session, gives you the token totals for the current conversation: input tokens, output tokens, and a rough dollar figure at API list prices.

This one is useful when you want to know whether a specific task blew through more tokens than expected. Long agent runs on Opus can burn tens of thousands of tokens in a single session; /cost shows the damage so you can decide whether to switch models, compact the context, or start a fresh session.

Between sessions, /cost resets. Claude Code doesn't maintain a persistent per-project counter across sessions in-app, which is why the historical methods below matter.

Method 3: Read the transcript files directly

Claude Code writes every session to disk as a JSONL file. The paths:

• macOS / Linux: ~/.claude/projects/<project-hash>/<session-id>.jsonl
• Windows: %USERPROFILE%\.claude\projects\<project-hash>\<session-id>.jsonl

Each project you've used Claude Code on gets a hashed directory; each session inside is one JSONL file. Every line in a file is one turn — a user message or an assistant response — and every assistant message includes token counts:

{
  "message": {
    "role": "assistant",
    "usage": {
      "input_tokens": 4823,
      "output_tokens": 512,
      "cache_read_input_tokens": 2100
    },
    "model": "claude-sonnet-4-6"
  },
  "timestamp": "2026-07-03T14:22:11.023Z"
}

A one-liner to see per-session totals across all your projects on macOS or Linux:

find ~/.claude/projects -name '*.jsonl' -exec \
  sh -c 'jq -s "[.[] | .message.usage.output_tokens // 0] | add" "$1" | \
  xargs -I {} echo "{} $1"' _ {} \; | sort -rn | head -20

That surfaces the 20 sessions that emitted the most output tokens. Useful for spotting the runaway session that burned through a chunk of your envelope.

The raw data is complete but the analysis is on you. If you want a rolled-up view without writing shell one-liners every time you're curious, that's what Method 4 is for.

Method 4: The free Claude usage dashboard

We built a Python-only, zero-dependency dashboard that reads the transcript files and generates a rolled-up view: daily, weekly, and monthly consumption, per-model breakdowns, per-project splits, API-equivalent dollar value, and a plan-optimizer verdict against Pro, Max, and API rates.

It runs locally — no account, no upload, no telemetry — and produces one HTML file next to itself.

The three commands:

git clone https://github.com/lucaspowell8020/claude-code-time-tracker
cd claude-code-time-tracker
python claude_usage.py --open --plan pro

Swap --plan pro for max_5x, max_20x, or api to match what you actually pay. The dashboard reads your transcripts, opens the report in your browser, and stops.

The full tool + explainer lives at /claude-usage. If you want to know which plan matches your actual usage before your next billing cycle, this is the shortest path.

What Anthropic's console does and doesn't show

A common confusion: the Anthropic console at console.anthropic.com shows API usage but not subscription-tier Claude Code usage.

If you use Claude Code with your Pro or Max subscription (the default), those sessions consume your subscription's rolling-window allocation and don't appear in the console. The console dashboard will look empty even after heavy Claude Code use.

If you use Claude Code in BYOK mode with an API key (export ANTHROPIC_API_KEY=sk-ant-...), the sessions consume API credits and do appear in the console. Same tool, different billing surface.

The transcript files record both cases because they log what Claude Code actually did, regardless of billing source. That's why they're the reliable source of truth.

Which method fits which question

Question	Best method
Am I about to hit a cap right now?	/status
How many tokens did this session burn?	/cost
Which project burned the most Opus?	Transcript files or dashboard
Should I upgrade to Max?	Dashboard (needs 30 days of history)
How many days last month did I hit limits?	Dashboard
What did my usage cost at API rates?	Dashboard
Do I appear in the Anthropic console?	Only if BYOK; otherwise no

A pattern worth stealing

The builders who stay ahead of Claude Code costs run the same loop: check /status a couple of times a day when working intensely, glance at /cost after a long agent run to spot the outliers, and run the dashboard once a week to see whether their tier is still right.

That last one is the highest-leverage step and the one most people skip. A 30-day view will surface things a single-session check can't: whether Opus is quietly draining your envelope, whether one project is doing 60% of your Claude spend, whether Pro's caps are costing you unblocked time you didn't realize.

If you want to start there, open the Claude Plan Optimizer — five minutes, no upload, no signup, and you'll know whether to keep, upgrade, or downgrade before you close the tab.