About a week ago, something unclear changed on Cursor’s side and Azure API stopped working in the API configuration for me.
Around the same time, Azure announced support for Claude Opus 4.5, and there were articles mentioning the integration — but it still seems delayed and not officially usable yet.
Instead of waiting, I referenced Cursor-Azure-GPT-5 and set up my own Azure Anthropic proxy deployed on Railway to keep using Azure + Claude inside Cursor.
If you’re facing the same issue or want to use Claude via Azure in Cursor, you can try this approach.
Azure Anthropic Proxy for Cursor
A lightweight proxy server that connects Cursor IDE to Azure Anthropic API (Claude) by exposing OpenAI-compatible endpoints.
Production URLs
-
Base URL:
https://cursor-azure-claude-proxy-production.up.railway.app/ -
Health Check:
https://cursor-azure-claude-proxy-production.up.railway.app/health
Supported Endpoints
-
POST /chat/completions— Main endpoint for Cursor (OpenAI format) -
POST /v1/chat/completions— OpenAI-compatible -
POST /v1/messages— Native Anthropic format -
GET /health— Server health check
How to use with Cursor IDE
-
Open Cursor Settings
-
Go to Models / Model Settings
-
Select Opus 4.5
-
Set OpenAI Custom API URL to:
https://cursor-azure-claude-proxy-production.up.railway.app -
Set OpenAI API Key to the same value as
SERVICE_API_KEYon the server
Important:
The API key in Cursor Settings must exactly match SERVICE_API_KEY.
If it doesn’t match, requests will be rejected with an authentication error.
Required Environment Variables
AZURE_ENDPOINT Azure Anthropic endpoint
AZURE_API_KEY Azure API key
SERVICE_API_KEY Auth key for Cursor requests
AZURE_DEPLOYMENT_NAME Default: claude-opus-4-5
PORT Default: 8080
Deployment (Railway – quick summary)
-
Deploy as a Node.js project
-
Set the environment variables above
-
Railway will auto-deploy and provide a public URL
-
Verify with:
GET /healthExpected response:
{ "status": "ok" }
License
MIT
Reference
Inspired by Cursor-Azure-GPT-5, a proxy service that enables Cursor to work with Azure GPT deployments.



