Cursor

Cursor with AI Gateway

Cursor is an AI-first code editor. Point its OpenAI API key settings at AI Gateway to reach every model in the gateway catalog and track spend in one place.

Set up with the Vercel CLI

Run the Vercel CLI setup command for Cursor:

terminal

npx vercel ai-gateway setup --agent cursor

The command provisions a key, exports it as AI_GATEWAY_API_KEY from your shell startup file, and prints the remaining setup steps with your values filled in. Cursor keeps API-key settings in its account-synced store instead of a file on disk, so you must finish the setup in Cursor.

Configuring Cursor

Complete these steps in Cursor after running the Vercel CLI. If you don't use the CLI, create an API key in the first step.

Get your API key
If you ran the Vercel CLI setup command on macOS, open a new terminal so your shell loads the AI_GATEWAY_API_KEY export, then copy the key without displaying it in the terminal:
terminal
```
printf %s "$AI_GATEWAY_API_KEY" | pbcopy
```
If you didn't use the Vercel CLI, go to the AI Gateway section in the Vercel dashboard sidebar and click API keys to create a key.
Open the model settings
In Cursor, open Settings with Cmd+Shift+J, then go to Models.
Add your key and base URL
Under OpenAI API Key, paste your AI Gateway key. Then enable Override OpenAI Base URL and set it to Cursor's dedicated compatibility endpoint:
```
https://ai-gateway.vercel.sh/cursor/v1
```
Cursor's base URL override sends /chat/completions bodies that the standard OpenAI schema rejects. Version 3.14.x sends flat tool definitions whose type isn't function, and newer builds send Responses-shaped bodies with input instead of messages. The /cursor/v1 endpoint normalizes both into standard Chat Completions. Every other path falls through to the standard /v1 handlers, so routing, billing, and errors are unchanged. Pointing Cursor at the bare /v1 or coding agent surface returns a 400 on tools.0.type instead.
Add the models you want
Use Add model to add gateway model IDs to the picker, for example anthropic/claude-opus-5 or openai/gpt-6-astra. Model IDs use the gateway's creator/model-name format, and the models catalog lists what's available.

Limitations

Cursor's bring-your-own-key support has constraints that come from Cursor itself, not from AI Gateway:

Limitation	What it means
Built-in models stop working	While the base URL override is on, Cursor's own non-OpenAI models are unavailable. Use gateway model IDs for everything, or turn the override off to go back
Tab completions are excluded	Tab never uses a custom key, so those requests don't reach the gateway
Agent and Auto may bypass the override	Some agent modes route to Cursor's own models regardless of the override
Requests pass through Cursor	Bring-your-own-key traffic still goes through Cursor's backend on the way to the gateway
No local sessions	Cursor stores chats in its backend, so there are no local sessions to migrate when you switch providers

Next steps

Browse the models catalog for available models
Configure spend limits on your API key
Connect your terminal agents with the Vercel CLI

Last updated September 13, 2026

Crush

Deep Agents CLI

Was this helpful?