Cursor is one of the few AI coding tools that reports usage at the event level — so you can see spend broken down by model and by team member, not just as one monthly number. StackSpend reads Cursor's usage events and rolls them into the AI Explorer alongside your API providers.
What Cursor exposes
Cursor's usage feed reports each event with the model used, the team member who made it, and — for token-based calls — a token breakdown split into input, output, cache-read, and cache-write. That's enough to answer the questions teams actually ask: which models are driving Cursor spend, which developers or teams are heaviest, and how the token mix looks per model.
StackSpend captures all of it in its Cursor usage monitoring dashboard: Cursor usage by model and by user, with the input/output/cache direction split on token-based calls, and a blended effective rate per model at your real mix.
The one limitation: token-based vs request-based calls
Cursor has two kinds of billed activity, and they carry different data:
- Token-based calls (Max mode and API-key usage) report the full token breakdown. These get the complete input/output/cache split in the AI Explorer, and expand in the per-model drill-down.
- Request-based calls (the classic included/fast-request pricing) carry no token breakdown — Cursor bills them per request, so there are simply no token counts to show. StackSpend counts them, but they have no direction split, and they sit outside the token totals rather than being padded with fake numbers.
This is a token-based vs request-based billing distinction, not a StackSpend gap — where Cursor reports tokens, you get them; where it bills per request, there's nothing to break down. The AI Explorer's coverage flags make that explicit rather than showing a misleading total.
What you can break Cursor spend down by
- Model — which models drive Cursor usage and cost
- User — usage per team member (Cursor has no separate project axis; the team is the account)
- Direction — input / output / cache-read / cache-write, on token-based calls
Related
- Cursor Usage Monitoring — Cursor usage per engineer, model mix, and seat activity in one dashboard
- AI Explorer — Cursor usage by model and user, next to every other provider
- Cursor Cost Monitoring — track Cursor spend across your team
- Model Recommendations — cheaper models at your real token mix
- How to Switch to Cheaper AI Models Without Losing Quality
FAQ
Can I see Cursor token usage by model?
Yes. Cursor exposes usage broken down by model and by user, which is enough to see which models your team runs and who is driving spend.
What is the limitation of Cursor's usage data?
Some calls are billed per request rather than per token, so a pure token view will not reconcile exactly to the invoice. Treat the token breakdown as an attribution tool rather than an accounting one.
How do I attribute Cursor spend to teams?
Cursor reports per user, so team attribution comes from mapping users to teams. That mapping has to be maintained outside Cursor, which is why it tends to drift.
