Budgethelm

Tracking Cursor usage and cost

Cursor is not a model vendor; you pay the price of whichever model you pick. This page covers what Budgethelm can and cannot see in Cursor.

Data as of 2026-10-31

Sign up free Log in

How Cursor cost accrues

Each chat or agent step in Cursor calls the model you chose, and the cost follows that model's token counts and prices. For Cursor's own plans and allowances, see Cursor's official documentation; this page does not describe them.

Model prices

Cursor lets you pick models from several vendors, so there is no single price table here. To compare a model, use the model price comparison, which ranks monthly cost for your usage.

Model price comparison and cost calculator。

How Budgethelm tracks Cursor

Cursor usage has three sources with different accuracy:

  • Local database (approximate): reads token counts from Cursor's local state.vscdb. The format is not publicly documented, and newer versions often store 0, which is ignored.
  • Hook character estimates: each tool call is estimated from character counts. The agent name gets "(estimated)" when uploaded, and the site lists them separately as "Token estimates", not in the exact figures.
  • Team Admin API (optional): the admin key stays on your computer and is only sent to api.cursor.com. Member emails are not uploaded.

Only time, tool, model, token counts and stop records are uploaded.

When Cursor gets stuck

Budgethelm adds a guard hook in ~/.cursor/hooks.json. When failures or repeated commands reach a threshold, it uses additional_context to ask the agent to stop (Cursor hooks have no output that ends the agent directly). This follows the official documentation but has not been run in a real logged-in Cursor, so no benchmark figures are shown.

FAQ

Can Budgethelm see exact Cursor spend?
Not always. The local database figures are approximate, hooks estimate from characters, and only the team Admin API gives official usage events. Each is labelled exact or estimated.
How much does Cursor cost per month?
This page does not answer Cursor plan pricing; see Cursor's official pages. Model API prices are in the model price comparison.
Does the guard work for Cursor?
The configuration follows the official documentation, but a stop has not been observed in a real Cursor session, so the honest status is "configured, not tested".