How to set a spending limit for Cursor & Claude Code

If you are a solo builder on a Cursor or Claude plan, you probably don’t have an admin API to pull usage from. Here’s a simple playbook that doesn’t require handing your login cookies to anyone.

Know which meter you’re on

• IDE / agent subscriptions (Cursor plans, Claude Code on Claude.ai plans): usage is shown in the vendor’s own dashboard. Cursor’s Admin API is for team admins (some endpoints, like per-user spend limits, are Enterprise only), so solo users shouldn’t expect a supported way for other tools to read their remaining credits.
• API $ (OpenAI/Anthropic organization billing): you can use the platform’s spend alerts and limits, and its cost/usage APIs if you have an organization admin key.
• Local session logs (Claude Code / Codex JSONL): tools like ccusage estimate spend from local log files. Great for history, but they only alert you if you run them yourself.

A budget that works without Enterprise APIs

1. Write down: plan price, reset date, and today’s remaining % (from the vendor UI).
2. Daily burn ≈ (used so far) ÷ (days elapsed). Days-to-empty ≈ remaining ÷ daily burn.
3. If days-to-empty < days until reset, cut speculative agent threads now.
4. Set calendar reminders to check mid-cycle and again around 80% used, so you don’t find out you’re empty in the middle of a deploy.

What not to do

• Don’t give a third party your account session cookies or password.
• Don’t upload raw agent JSONL that contains prompts/source—if you sync, send aggregates only (daily totals by model/project).
• Don’t assume OpenAI API billing covers your ChatGPT or Codex subscription usage. They are billed separately.

If you do have OpenAI API spend

Use official spend alerts plus optional hard limits. Alerts notify; hard limits 429 traffic. Guide: OpenAI Spend limits. Companion page: alerts vs hard limits.