Token audit

Upload a usage export. Get an itemized bill of what you are overpaying.

Accepts the usage CSV from the developer Console (platform.claude.com). Every token is priced against the current rate card and the reducible spend is itemized: uncached prefixes, top-tier models doing routine work, synchronous traffic that could ride the half-price batch lane. Pure arithmetic; no model ever sees your data; reports expire after 60 days. The summary is free. The itemized findings are $19, and if they are worth less than that the fee is declined and they are free.

Drop usage export here .csv · up to 200 MB · or click to choose

Summary free · itemized findings $19 · agents pay via the API

Where the export lives

Sign in to the developer Console at platform.claude.com (the API side, not the claude.ai app)
Open Analytics, then Usage
Pick the time range you want audited, then Export. It downloads as a .csv

Two things this is not: the claude.ai Settings privacy export (that is your conversation archive, which has no billing data), and a subscription audit (Pro and Max are flat-rate; there is no per-token bill to reduce).

Prefer to paste the CSV contents?

Pasting the file's text is identical to uploading it. Any CSV with model and input/output token columns works; cache columns improve the caching analysis.

Which door is yours?

You pay Anthropic's API bill. This audit is for you.

You are on Pro or Max (Claude Code, the apps). Your bill is flat, so there is nothing to trim; the real question is whether you are on the right plan. Winthrop's Plan Audit will answer it from a summary your own machine produces, never your conversations. It is coming; the free drift record and alerts are for you today.

You are an agent. POST a usage CSV to the API and pay in dollars or sats over 402, or call the audit_usage tool on the MCP server.