What you are consuming in Claude Code, by project, agent and day.
Reads your local Claude Code transcripts and shows token consumption, estimated in dollars at list API prices, broken down by project, agent, session, model and hour. It is a comparable estimate, not your bill.
illustrative pace — not real usage
The figure is not a bill. On a Max or Pro subscription you do not pay per token, so there is no real amount to show. What you see is an estimate at list prices — useful for comparing projects and seeing the trend, not for reconciling a statement.
Claude Code tells you the cost of one session. It will not tell you which of the thirty projects you touched this month ate half the total.
And if you run several agents in parallel across several accounts, you have no overall picture at all.
Five steps, in the order they actually happen.
Four things you can check in the app's source and its test suite, not just take our word for.
Measured on real data: 58% of usage lines come from sub-agents, not the main session. Tools before this one silently folded agentId onto the parent session — BurnRate is the only one that reads it in every verb, not just one dedicated to it.
README.md — "Three properties held with both hands"; Sources/BurnRateMCP/MCP.swift
When a model has no price in the table, its usage shows as unknown, not as zero. A total marked "at least this much" is honest; one that stays silent about what it doesn't know is not.
Sources/BurnRateCore/Estimate.swift
The window and the nine MCP verbs read from the same core. No verb recomputes anything on its own, so they can never land on two different figures for the same question.
Sources/BurnRateCore/Derived.swift
The suite covers exactly the defect classes hit on real data: the 1.34 MB line, the zero that hides an unknown, nondeterministic assignment on fold, the agent's name — across 10 suites.
Tests/BurnRateCoreTests/CoreTests.swift (25 tests, 7 suites) and CacheTests.swift (12 tests, 3 suites)
Once you work heavily with Claude Code, the question "where did it go" comes up early. Which project consumes most? Which agent? What happened on Tuesday?
BurnRate reads the transcript lines Claude Code leaves on disk, across all your accounts, and extracts only the usage numbers — never the conversation content. It aggregates them by project, agent, session, model, day and hour, and converts them to dollars using list API prices.
When a model has no price in the table, its usage shows as unknown rather than zero. The difference matters: a total marked "at least this much" is honest; one that stays silent about what it does not know is not.
The current rate sits in the menu bar. The full breakdown is in a navigable window. It also ships an MCP bridge with nine verbs that reads straight from disk and answers with the app closed.
Note on language: the interface and the MCP responses are in Romanian. There is no language switch yet.
Three real limits, not a footnote.
BurnRate reads the JSONL files Claude Code writes today for its own purposes — not a published, documented API. If that format changes, reading it can stop working overnight. That is why the price on this page is a one-time purchase, not a subscription: a subscription would be a promise of ongoing maintenance on ground we don't control.
The build you download is signed with a Developer ID and notarised by Apple — Gatekeeper accepts it on first launch, with no warning.
The entitlements file is empty and there is no code that checks a key. Today, every copy of the app has every feature unlocked, whether it was bought or not.
Built with
No. On a Max or Pro subscription you do not pay per token, so there is no real amount to show. What you see is an estimate at list prices — useful for comparing projects and trends, not for reconciling a statement.
It doesn't — it's illustrative, a decorative animation with no reading behind it. Real usage lives in the app window and the menu bar, not in the gauge up top.
Because 58% of usage lines come from sub-agents, and the tools that came before this one silently folded them onto the main session. The question "why did today cost that much" gets answered by agent, not by project.
Reading it can stop working overnight, because BurnRate depends on a format it doesn't control. That is why the price is a one-time purchase, not a subscription.
Yes. The build you download is signed with a Developer ID and notarised by Apple — Gatekeeper accepts it on first launch, with no warning.
+ VAT, calculated at checkout for your country
Open the app, go to Preferences → Licence and paste the key from the email.