Please clarify the Custom API Keys dialog and settings

When attempting to set up a custom API key, Cursor displays this prompt:

Are you sure you want to enable your own ${provider} API key? Several of Cursor’s features require custom models (Tab, Apply from Chat, Composer), which cannot be billed to an API key.

The implication at this point is unclear:

  1. Will features that require custom models not work? OR
  2. Will those features work, but Cursor will still charge for them?

Proposed language if #1 is the case:

Are you sure you want to enable your own ${provider} API key? Several of Cursor’s features require our custom models (Tab, Apply from Chat, Composer), so they will be disabled.

Proposed language if #2 is the case:

Are you sure you want to enable your own ${provider} API key? Several of Cursor’s features require our custom models (Tab, Apply from Chat, Composer), and they will be billed separately from your API key.

To further clear up the confusion (see also this post), adding custom API keys should be warned against, or even disabled from the start, for Pro users. I’m saying this because after enabling my Anthropic API key, I see this:

With your Cursor Pro subscription, you do not need to use your own Anthropic key!

If I don’t need to use my own key, then why let me enter it? Warn me before.

…if it’s really true that I don’t need to enter my own API key. Because at cursor.com/settings I see this:

Enable for usage-based for premium models :white_question_mark:
“Applies to models like Sonnet 3.5 and GPT-4o beyond the monthly requests included in your plan.”

So it seems that it does make sense to enter my own Anthropic API key, even if I have a Cursor Pro subscription, in order for Cursor to not charge me for premium model usage.

CC @litecode @lonelyplanetdev @plymouthvan

8 Likes

Have you gotten an answer on this? I have the same question.

1 Like

+1. It’s still unclear whether the implication is one or the other?

+1. I am also unclear on this.

+1. This is really annoying. I’m not going to use my Anthropic key if it will prevent other features from working as well, but the language makes it impossible to know what will happen.

I decided to do some experimenting to discover what would happen. Looks like there are four ways to get billed for claude here. (can’t speak for the other models exactly)

  1. “claude code” via the button
    this generates a new api key via oath and bills you normally outside of cursor (same as 2)
  2. claude command via pure terminal window, which seems to have a clumsy integration with the ide, but it does work, just not as fluid as 3
  3. “agent” chat with claude model selected (because auto wastes human mental cycles)
    this uses the cursor subscription and is limited to 200k context window
  4. use your own key in settings; not really a new way to use claude, but a new way to get billed maybe? seems to combine 2+3, bring your own key, tabs still work
    seems ideal for claude cause we have 1m context at work, so i’d rather use that then the one given by cursor