Claude Code spend analyzer
Drop your ~/.claude/projects folder and see what your Claude Code usage is worth, broken down by project, git branch, and model. The analyzer reads the same session logs Claude Code already writes, the same way BudgetClaw does, and it analyzes them without uploading anything.
Runs entirely in your browser. Nothing you enter leaves this page.
Your logs never leave your browser. No upload, no account. These files contain your prompts; only token counts and model ids are read.
Choose your ~/.claude/projects folder
or drop .jsonl session files here. Parsed locally, never uploaded.
Where your logs live
- Claude Code writes one .jsonl file per session under ~/.claude/projects, grouped by project.
- Click the box above and choose that projects folder (your browser asks permission to read it locally).
- Everything is parsed in the tab. Nothing is uploaded, and prompt text is never read.
Engine verified on 2026-07-25: it reproduces BudgetClaw per-project totals to the cent on real logs. Cost is usage value at Anthropic list rates, not a subscription bill.
Frequently asked questions
Is this my actual Claude bill?
It is the dollar value of your token usage at Anthropic's published API list rates. If you code through a Claude Pro or Max subscription, you pay a flat monthly fee, so this is the value your plan covers, not an extra charge. If you run Claude Code on an API key, it is your actual metered cost. Either way it answers "where did the usage go", broken down by project and branch.
Does anything get uploaded? These logs contain my prompts.
Nothing is uploaded. The files are read and parsed with JavaScript inside your browser tab. There is no upload endpoint and no analytics event that contains your data. You can confirm in your browser dev tools, or disconnect from the network after the page loads and it still works. Only token counts and model ids are read; prompt text is never used.
How is the cost calculated?
Per model, from the token counts Claude Code records: input and output tokens at the model rate, cache reads at 0.1x input, and cache writes at 1.25x input (5-minute) or 2x (1-hour). Identical API responses written across several log lines are de-duplicated on message id plus request id, the same way they should be, so nothing is double counted.
Where do I find the folder to drop?
Claude Code writes session logs to ~/.claude/projects on your machine, one .jsonl file per session, organized by project. Choose that folder above (or drop the .jsonl files directly). Everything stays local.
Why might this differ from what Anthropic shows?
Subscription plans do not bill per token, so there is no per-token line to match. For API-key usage, small differences come from list-rate rounding and any rate changes since the verified date. The figures here use standard list rates verified on the date shown below the tool.
Can I share a screenshot without leaking client or project names?
Yes. Turn on "Anonymize names for screenshots" above the results and every project and branch label is replaced with a neutral alias (Project 1, branch-1, and so on), while the dollar, token, and share values stay exactly as shown. Nothing is sent anywhere either way; this only changes what the table displays.
Sources
Prices last verified 2026-07-25, from each provider's official page.
Independent tool, not affiliated with Anthropic. Usage-value estimates at list rates; your subscription or API bill is set by Anthropic.
This is one snapshot of one machine. Keep the history, watch every project and provider, and get budget alerts in one dashboard. See the dashboard
Want your whole stack in one number, subscriptions included? Total it with the AI stack cost auditor.