Cursor
Last updated August 26, 2026
Point Cursor's custom OpenAI base URL at Amux, with notes on which models currently work.
Cursor lets you override the OpenAI address and key, so point them at Amux.
A Cursor Pro subscription is required — custom models and third-party APIs are a Pro feature.
1. Create an Amux key
Create one on the keys page in the console. The key remains viewable there after creation, so you do not need to rotate it just to retrieve it again.
2. Configure
- Open Cursor Settings (the gear) and pick Models on the left
- Find OpenAI API Key: turn the toggle on and paste your Amux key
- Find Override OpenAI Base URL: turn the toggle on and enter
https://gateway.amux.ai/v1 - Click + Add Custom Model, enter a canonical ID (for example
anthropic/claude-opus-5), and make sure its toggle is on (green)
3. Check it works
Press Cmd + L (macOS) or Ctrl + L (Windows) to open a chat, pick the model you just added, and send a message.
Known limitations
These two limitations come from Cursor's current custom-OpenAI implementation and can affect Amux integrations as well. Cursor treats a custom address as if it were OpenAI's own service, but its request shape differs from standard Chat Completions in some cases.
| Models | Symptom | Status |
|---|---|---|
| GPT family | Parameter messages is required | Cursor sends these in the Responses request shape while the custom-address field routes to Chat Completions, and the two do not line up. No workaround today |
| Claude family | Tool-definition errors in some modes | Cursor sends tool definitions along with the request in a shape that is not fully compatible with Chat Completions |
Both issues depend on changes on Cursor's side. If either is a blocker, Cline uses the standard OpenAI-compatible protocol and does not have these limitations.
Choosing models
Every model you want has to be added under Add Custom Model — this field does not fetch a model list. Names are canonical IDs.
To pin a provider or choose a routing strategy, enter the full suffixed form (anthropic/claude-opus-5:@price) — see the API reference.
Troubleshooting
The model I added is not in the dropdown
Go back to the Models page and check that its toggle is green. Newly added models may default to off.
Authentication failure
Both toggles have to be on: filling in the key without enabling Override OpenAI Base URL still sends the request to OpenAI itself, where an Amux key will fail.
Tab completion still uses Cursor's own model
The custom OpenAI address only applies to chat. Code completion still goes through Cursor's own service, which is a product limitation rather than a configuration issue.