A Free Claude Code Cost Tracker

“Tracking Claude Code cost” sounds like it should be trivial — there's a /usage command right there in the CLI. The problem is that /usage only shows your current session window; it doesn't accumulate history across sessions, doesn't show you daily or weekly trends, and gives you no way to ask “how much did I spend in the last 30 days?” — which is the question that actually matters for budgeting.

What “tracking” actually means

A useful Claude Code cost tracker needs to answer four questions: How many tokens did I burn over a period (day, week, month)? Which model did I use most? Is my subscription saving me money compared to API rates? And what does my trend look like — am I burning more or less than last month?

The raw data to answer all four questions is already on your machine. Claude Code writes a detailed JSON usage log for every session, stored under ~/.claude/usage/. The only thing missing is a reader that aggregates across files and converts tokens to dollars.

Setting up your tracker in 30 seconds

There's nothing to configure. The tracker reads whatever logs exist and returns a breakdown immediately. You don't need to “start” tracking — if you've used Claude Code at all, the data is already there waiting.

$ npx whoburnedmore↳ found 47 usage sessions across 31 days   CLAUDE CODE COST TRACKER  (last 30 days)  week           sessions   tokens       est. cost  ─────────────────────────────────────────────────  Jun 9–15       14 sess.   122.4M       $26.30  Jun 2–8        11 sess.    88.1M       $19.00  May 26–Jun 1    9 sess.    61.2M       $13.20  May 19–25      13 sess.    94.7M       $20.40  ─────────────────────────────────────────────────  30-day total   47 sess.   366.4M       $78.90   ↳ Claude Max ($100/mo): saving ~$21 this month 🔥

1. 1

   Run the tracker

   npx whoburnedmore in any terminal. It reads your local logs without network access.
2. 2

   Check the weekly trend

   Look at whether cost is increasing week-over-week. A rising trend usually means longer sessions with bigger context — not more sessions.
3. 3

   See the model split

   If your usage is split between Sonnet and Haiku, the breakdown shows cost per model. Heavy Sonnet usage on simple tasks is usually the biggest savings opportunity.
4. 4

   Share your leaderboard entry (optional)

   Run without --local to upload your token totals (not your code) and appear on the whoburnedmore public leaderboard. You can share the link or keep your dashboard private.

How it compares to tracking manually

Developers who track cost manually usually do one of three things: copy numbers from/usageinto a spreadsheet after each session, check the Anthropic Console billing page periodically, or eyeball their credit card statement. All three approaches have the same flaw: they're reactive rather than proactive, and none of them give you session-level detail alongside a monthly total.

Flat-plan users: why the Console doesn't help you

If you're on Claude Pro or Max, the Anthropic Console billing page shows one number: your flat subscription charge. It doesn't show you how many tokens you burned or whether you're getting value for your subscription fee. Whoburnedmore fills that gap by computing the API-equivalent cost of your actual usage and comparing it to what you pay.

Privacy: what stays local and what doesn't

Whoburnedmore is designed for developers who are cautious about what they send upstream. The default behavior sends only aggregate token counts and timestamps to the leaderboard — your code, prompts, and file names never leave your machine. If even that is too much,--local mode opens a browser dashboard directly from your local files without any network request.

What each mode does

$ npx whoburnedmore↳ uploads aggregate token totals, joins leaderboard $ npx whoburnedmore --local↳ opens browser dashboard, nothing sent upstream $ npx whoburnedmore --dry-run↳ prints what would be submitted, sends nothing

The shareable dashboard

When you run whoburnedmore without --local, you get a shareable URL for your personal dashboard — a public page showing your token burn and leaderboard rank. Some developers share this with their team to make AI usage visible; others share it socially to compare with other developers. The data shown is always aggregate (token counts and dates), never session content.

The best cost tracker is one you actually use. Whoburnedmore requires no setup and takes under 10 seconds to run, which means it's low enough friction to become a genuine weekly habit. Run npx whoburnedmore right now and see where you stand — the logs have been accumulating since your first Claude Code session. 🔥