Claude Code Usage: How to See What You're Actually Using
You can't manage what you can't see — and Claude Code usage is, by default, almost invisible. You fire off prompts, agents read files, sessions run for hours, and the only signal you get is a vague sense of dread or a rate-limit wall you hit out of nowhere.
This guide covers exactly how to see your Claude Code usage: the built-in command that shows it, the tools that chart it over time, how the usage limits on Pro and Max actually work, and how to keep an eye on all of it without thinking about it.
What "usage" means in Claude Code
"Usage" gets blurry because there are two different meters, depending on how you pay (we break the billing down fully in Claude Code cost: what you actually pay):
- On a subscription (Pro / Max), usage is measured against an allowance with rate limits. You don't see dollars — you see how close you are to a ceiling.
- On an API key, usage is measured in tokens — input, output, and cache reads/writes — each priced per model, billed as you go.
Either way, the underlying unit is the same: tokens. Every message re-sends your conversation plus whatever files are in context, so usage climbs with session length and context size, not just with how many prompts you type.
Check your usage right now: the /cost command
Claude Code has a built-in way to see what the current session has used. In an active session, run:
`` /cost ``
It reports the tokens and (on API billing) the dollar cost for the current session. It's the fastest gut-check there is — but it has two limits: it's per session (close the terminal and it's gone), and on a Pro/Max subscription it reflects allowance usage rather than a real dollar figure.
So /cost answers "what did this session cost?" It does not answer "what have I spent this week?" or "which project is eating my budget?" For that you need history.
See historical usage: ccusage
Claude Code writes detailed logs of every session to your machine. The most popular way to turn those into a usage report is ccusage (★ 16.1k), an open-source CLI that parses those local JSONL logs and prints daily, monthly, and per-session breakdowns:
`` npx ccusage@latest ``
It gives you token counts and estimated cost per day and per model, entirely from local data — nothing is uploaded. It's excellent for an on-demand "how did last week look?" report.
The gap: it's a command you have to remember to run. It's a rear-view mirror, not a live gauge — which is the whole problem with usage you only check after the bill scares you.
How Claude Code usage limits work
If you're on a subscription, the usage that matters isn't dollars — it's whether you're about to hit a wall:
- Rate limits cap how much you can do in a rolling window. Hit them and Claude Code makes you wait before continuing.
- Plan tier changes the ceiling. Max raises the allowance substantially over Pro, which is the entire reason to upgrade — we compare the tiers in Claude Code Pro vs Max.
- Heavy agentic work burns the allowance fastest. Long sessions, subagents, and large file reads consume far more than a quick chat, so power users hit limits they didn't expect.
The frustrating part is that the limit arrives without warning. Knowing roughly how hard you've been pushing before you hit the ceiling is the difference between pacing yourself and losing an hour mid-task.
Monitor usage continuously (so you stop getting surprised)
The tools above are each missing the same thing: they're things you check, not something that's just there. /cost is per-session; ccusage is an on-demand report. Nothing sits in front of you while you work.
That's the gap Tokipet fills. It's a macOS menu-bar app that reads the same local Claude Code logs ccusage does, and keeps a live usage and spend readout in your menu bar — today's tokens, the month so far, and a breakdown by model — updated as you work. It's local-first: your logs never leave your Mac, and it's free during the beta. (It turns the number into a collectible pixel pet too, which is a far better motivator to actually watch your usage than a spreadsheet.)
If you want a usage dashboard you don't have to remember to open, that's the one to grab. If you'd rather stay in the terminal, ccusage is the right call.
How to use less (not just see more)
Seeing your usage is step one; bringing it down is step two. The biggest levers are context size, session length, and model choice — we cover all nine in how to reduce Claude Code token usage. The short version: keep context lean, start fresh sessions often, and don't run Opus for work Haiku could do.
FAQ
How do I check my Claude Code usage? Run /cost inside a session for the current session's tokens and cost. For history across days and projects, run npx ccusage@latest, or use a menu-bar monitor like Tokipet that keeps the numbers live.
Does Claude Code show token usage? Yes — /cost shows the current session's token usage, and Claude Code logs every session locally so tools can report token usage over time.
What is the Claude Code usage limit? On Pro and Max, usage is capped by rate limits over a rolling window; Max has a much higher ceiling than Pro. On an API key there's no fixed limit — you pay per token for whatever you use. See Pro vs Max for how the tiers compare.
Can I see usage per project or per model? Not from /cost alone. ccusage breaks usage down by model, and Tokipet shows spend per model and per repo from the same local logs.
Why is my Claude Code usage so high? Usually context size and long sessions: every turn re-sends the whole conversation plus files in context, so cost compounds within a session. Reducing token usage walks through the fixes.
Tokipet is a free, local-first macOS menu-bar app that turns your Claude Code usage into a live dashboard — and a pixel pet. Download it free and stop guessing what you're spending.