TokenLoop
Stop AI coding-agent spend before the invoice.
Spend control for Claude Code and Cursor: per-developer burn, a minutes-scale kill switch, and client chargeback — without proxying traffic through a gateway. Free signup; keys encrypted at rest. Since flat-rate plans gave way to routed-model pricing, the same agent loop bills a different amount depending on which model answered.
- Org burn$514.10
- Daily cap$250
- jordan@ (spike)$410.60 · 7.8×
- Kill switchRevoke requested
- Chargeback MTD$1,180 bill-back
What it does
- Per-developer Claude Code + Cursor spend in one pane
- Kill switch: detect, alert, throttle, or revoke (admin APIs)
- Agency chargeback with markup and CSV export
- Keys encrypted at rest. Never returned to the browser
- Honest scope: minutes-scale cut, not true per-request blocking
How it works
- 1Create a free account (email + password).
- 2Paste Anthropic and/or Cursor admin keys. We AES-256-GCM encrypt them before storage.
- 3Sync spend, set a daily cap / spike multiplier, and optionally tag clients for chargeback.
- 4Kill switch detects spikes ~every 10 minutes and can throttle or revoke via admin APIs.
Who it’s for
Engineering and agency leads running Claude Code and Cursor who find out about runaway spend on the invoice — now that enterprise flat rates are expiring and billing follows whichever model the router picked. They need detect-and-cut plus client bill-back without a proxy.
What’s next
- Spend dashboard + sync — shipped
- Kill switch + chargeback — shipped
- Free signup · all features — shipped
- Codex / OpenAI ingestion
Need a human to read this report?
The report is the start. Studio turns findings into a plan with owners. Thirty minutes to see if that's worth it.