I’m posting this to warn others and demand accountability from Cursor.
On July 15, my Pro plan renewed, giving me the $20 monthly quota that includes usage of the Claude 4 Opus model. I opened Cursor and submitted a single prompt.
Without any further action from me, the agent system recursively triggered nearly 50 separate Opus calls in under 10 minutes, each burning massive token counts (~100K–150K+). All were silently logged as “Included in Pro” and counted against my usage.
I was immediately hit with “You’ve hit your usage limit” after only a few manual actions. Reviewing the log confirmed the truth:
One prompt = 50 background model calls = total plan drained instantly.
There was:
No confirmation
No warning
No limit on recursive agent behavior
No visibility into how fast my budget was being obliterated
This is unacceptable.
Cursor promotes agent-based workflows, yet punishes users for using them. If agents are allowed to run wild and spend your plan balance autonomously, this must be opt-in, transparent, and capped.
Otherwise, you’re selling a tool that sets users up to burn through their quota without knowing what’s happening.
I’ve contacted support to demand a reset or refund. But the fact that this is even possible is the real issue. It needs to be fixed immediately.
If you’re on Pro and using Opus with agents, check your logs. You might be getting drained without realizing it.
I just got a response from cursor support, draw your own conclusions:
Hi Illarion!
You’re absolutely right - I can see that a single agent prompt triggered multiple recursive Opus requests, quickly consuming your monthly budget. This shouldn’t happen without warning, and I understand your frustration.
Unfortunately, we can’t reset usage quotas since they represent actual computational resources used. However, I can help you with two options:
Cancel your current subscription and request a refund of this month’s payment
Continue using Cursor by switching to Auto mode (unlimited usage) or enabling usage-based pricing with a spend limit you set
Let me know which option you’d prefer and I’ll help you proceed.
Refund means I have to cancel my subscription and loose all my chats. Not an option. I don’t think it matters what plan you are on, if you put in a request and leave, it’s possible that when you come back your account is completely drained because the model went off on a tangent. This is a serious problem and I can’t imagine I am the first to experience it.
I’m a simple guy, if I see BS, I speak up. If I see something great, I speak up. Cursor is expensive as hell, but it handles better than any other wrapper (Cline, Codeium, etc.). It’s the best application out there, just an expensive middle-man mfer.
I’ve noticed that Cursor is truly misleading customers. Since the beginning of the month, I haven’t made that many requests compared to usual, yet I’m still seeing this message. It’s really hard to accept the lack of transparency in their pricing.
It is not your usage. The model goes off and makes its own decisions whether you ask it or not but they charge you for it. It’s as simple as that. You are given no warning of any kind. You put in a request and go to the bathroom, by the time you come back you are down 50 requests. Granted my bathroom break could be long but still
Does it matter what the plan is? Are you okey with paying $20 for 1 request? You are also assuming that the model went off to work on what I asked it to do. This was not the case. It started making changes where I did not ask it to.
Dude the model got stuck recursing in background agent. What are you not understanding? That is not intended usage that’s unreal that this is able to happen. Cursor has become a complete shitshow since the start of July.
Unless we know what exactly happened it is useless to talk about it. I give cursor a single feature request, it does a few calls to implement it, and does a few more to bug fix. Yea sometimes 50 or more. If your prompt isn’t extremely specific it is going to do more than you want it to. OP doesn’t understand how cursor works so until he gives better info it is best to assume user error.
I get where you’re coming from, but this wasn’t a case of me not knowing how Cursor works.
I have the logs. One prompt — not vague, not open-ended — triggered over 50 Claude Opus calls in under 10 minutes. I didn’t keep clicking, didn’t loop it, didn’t touch it after the first request. Cursor support already confirmed it:
“You’re absolutely right – I can see that a single agent prompt triggered multiple recursive Opus requests, quickly consuming your monthly budget. This shouldn’t happen without warning.”
So no, this wasn’t user error. This was the agent system going off on its own and burning my entire monthly quota without telling me or giving me a way to stop it.
The issue isn’t that Cursor made a bunch of calls. The issue is that it did it on its own and charged me for all of it, silently. That’s the problem — not how specific my prompt was.
I’m not mad at Cursor trying to do something powerful — I just don’t think users should get punished for using the tool the way it’s designed.
