Complaint: Inability to Use Personal API Keys in Cursor 0.47 - Pro User Frustration

I am writing to express my frustration and disappointment with the recent changes in Cursor 0.47, specifically regarding the inability to use personal API keys. As a Pro user, I believe this change significantly diminishes the value and flexibility of Cursor.

Here are the key issues I am encountering:

  1. Loss of Custom API Key Functionality: Previously, I was able to seamlessly use my own API keys with Cursor. This was a crucial feature for me as a Pro user. Often, for simpler tasks, models like the free DeepSeek R1 and Gemini are perfectly adequate. Using my own API keys for these models allowed me to reserve my Pro requests for more demanding tasks and optimize my resource usage. Now, this option seems to be completely removed.

  2. OpenAI/OpenRouter API Key Validation Error: When I attempt to use my OpenAI API key, or even try to use an OpenRouter key, the system throws a validation error. I have attached a screenshot below for your reference. This indicates a clear issue with the API key verification process.

  1. Google Gemini API Key Validation Issue: Similarly, when I try to input my Google Gemini API key, it gets stuck in the “Verifying Gemini 1.0 Pro API key…” state and ultimately fails to validate. This makes it impossible to utilize my own Gemini API key within Cursor.

As a result of these issues, I am currently forced to rely solely on the models provided by Cursor, and I am unable to leverage my own API keys for any models. This raises a significant question: Why has this functionality been removed?

Could this change be intentionally implemented to force Pro users to consume more of Cursor’s official requests, even when using models that could be powered by our own API keys? This feels like a step backward in terms of user control and customization.

I urge the Cursor team to address these issues promptly and restore the functionality to use personal API keys. Understanding the reasoning behind this change and finding a solution is critical for maintaining user satisfaction, especially for Pro users who rely on Cursor for efficient and flexible workflows.

4 Likes

Honestly, I’m surprised that a simple quick fix to specify a custom model to use in order to test an api endpoint has still not been urgently rolled out yet after a week of documenting this issue (which from the responses of 0.47 so far has been concerning this issue). At least for the quickest patch, have the user be able to select a model from the model list to test the api or just save the key regardless (make key override available in settings):

Hey, this will be fixed soon.

Seems Cursor “pushes” to use it’s Models and Pro mode.
Hate thism esp. when Gemini 2.5 Pro released and Deepseek new build rolled out!!!

1 Like

This is still not working.