Skip to content
BranchesTags
 
 

Repository files navigation

Azure Anthropic Proxy for Cursor

Proxy server để kết nối Cursor IDE với Azure Anthropic API (Claude).

🌐 Production URLs

📋 Endpoints

Root Endpoint

  • GET / - Thông tin về server và các endpoints có sẵn

Health Check

  • GET /health - Kiểm tra trạng thái server

Chat Endpoints

  • POST /chat/completions - Endpoint chính cho Cursor IDE (OpenAI format)
  • POST /v1/chat/completions - OpenAI format
  • POST /v1/messages - Anthropic native format

🚀 Cách sử dụng

Cấu hình trong Cursor IDE

  1. Mở Cursor Settings
  2. Tìm phần "Model" hoặc "Model Settings" Mở "Opus 4.5"
  3. API Keys mucj OpenAI Custom API URL: https://cursor-azure-claude-proxy-production.up.railway.app
  4. Đặt API Key: Giá trị phải trùng khớp chính xác với biến môi trường SERVICE_API_KEY trong file .env của server. Bật OpenAI API key

Cấu hình Model trong Cursor IDE

Cấu hình Chat trong Cursor IDE

Lưu ý quan trọng: API Key trong Cursor IDE (Cursor Settings > Models > API Keys > OpenAI API Key) phải khớp chính xác với giá trị SERVICE_API_KEY trong file .env của server. Nếu không khớp, request sẽ bị từ chối với lỗi authentication.

Ví dụ Request

curl -X POST https://cursor-azure-claude-proxy-production.up.railway.app/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_SERVICE_API_KEY" \
  -d '{
    "model": "claude-opus-4-5",
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Lưu ý: Thay YOUR_SERVICE_API_KEY bằng giá trị thực từ biến môi trường SERVICE_API_KEY.

⚙️ Environment Variables

Claude (Azure Anthropic)

  • AZURE_ENDPOINT - Foundry resource base URL, e.g. https://<resource>.services.ai.azure.com (bare host, no path). The server appends /anthropic/v1/messages for Claude and reuses the same host for the Azure OpenAI calls. The legacy full-URL form (.../anthropic/v1/messages) is still accepted for backward compatibility.
  • AZURE_API_KEY - Azure API key (shared with the OpenAI endpoint on the same Foundry resource)
  • AZURE_CLAUDE_DEPLOYMENT_NAME - Default Claude deployment name, e.g. claude-opus-4-7. Used for every Claude request unless a family-specific override is set.
  • (optional) AZURE_CLAUDE_OPUS_DEPLOYMENT, AZURE_CLAUDE_SONNET_DEPLOYMENT, AZURE_CLAUDE_HAIKU_DEPLOYMENT - override the deployment per model family

GPT-5.4 (Azure OpenAI)

Azure OpenAI uses a different host than Anthropic (cognitiveservices.azure.com instead of services.ai.azure.com), so AZURE_OPENAI_ENDPOINT is required for GPT. Leave it unset to disable the GPT route — Claude keeps working either way.

  • AZURE_OPENAI_ENDPOINT - Azure OpenAI endpoint. Two forms are supported:
    • v1 API (recommended): https://<resource>.cognitiveservices.azure.com/openai/v1
      • Request URL becomes <endpoint>/chat/completions?api-version=preview — the api-version is fixed and AZURE_OPENAI_API_VERSION is ignored.
    • Legacy deployment URL: https://<resource>.cognitiveservices.azure.com
      • Request URL becomes <host>/openai/deployments/<deployment>/chat/completions?api-version=<AZURE_OPENAI_API_VERSION>.
  • AZURE_GPT_DEPLOYMENT - Exact name of the gpt-5.4 deployment in your Azure resource. Default gpt-5.4
  • AZURE_OPENAI_API_VERSION - Only used with the legacy endpoint form. Default 2025-04-01-preview.
  • GPT_REASONING_EFFORT - Fallback effort (minimal / low / medium / high) when Cursor sends the bare gpt-5.4 id without a suffix. Default medium
  • (optional) AZURE_OPENAI_API_KEY - defaults to AZURE_API_KEY

Service

  • SERVICE_API_KEY - Key used to authenticate Cursor against this proxy (must match Cursor's OpenAI API Key field)
  • PORT - Server port (default 8080; Railway assigns its own)
  • (optional) LOG_TOOL_CALLS, LOG_MESSAGES - verbose logging toggles (true/false)

🤖 Using GPT-5.4 in Cursor

Azure Foundry ships gpt-5.4 with native Chat Completions support since 2026-03-05, so the proxy forwards Cursor's OpenAI request almost unchanged (only swapping the model name to the deployment, stripping temperature/top_p, and converting max_tokens → max_completion_tokens, as required by reasoning models).

  1. Deploy gpt-5.4 in the same Foundry resource you use for Claude. Note that gpt-5.4 and gpt-5.4-pro currently require access registration on Azure.

  2. Set the AZURE_OPENAI_* env vars above on Railway and redeploy.

  3. In Cursor Settings → Models → Custom Models, add any of these model ids. Each one maps to the same Azure deployment, only reasoning_effort changes:

    Cursor model id reasoning_effort
    gpt-5.4 falls back to GPT_REASONING_EFFORT (default medium)
    gpt-5.4-minimal minimal
    gpt-5.4-low low
    gpt-5.4-medium medium
    gpt-5.4-high high

  4. Keep the OpenAI API Key field set to your SERVICE_API_KEY and the base URL pointing at this proxy — both Claude and GPT-5.4 share the same proxy endpoint and auth key.

Quick smoke test

curl -X POST https://cursor-azure-claude-proxy-production.up.railway.app/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SERVICE_API_KEY" \
  -d '{
    "model": "gpt-5.4-medium",
    "messages": [{"role": "user", "content": "Say hi in one word."}]
  }'

📦 Installation

npm install
npm start

🔧 Development

npm run dev

🚂 Deploy trên Railway

Cấu hình nhanh

  1. Tạo project mới trên Railway

    • Truy cập Railway
    • Tạo project mới từ GitHub repository hoặc Deploy từ GitHub

  2. Cấu hình Environment Variables

    • Vào tab Variables trong Railway project
    • Thêm các biến môi trường sau: 
      # Azure Anthropic
AZURE_ENDPOINT=https://<resource>.services.ai.azure.com
AZURE_API_KEY=your-azure-api-key
AZURE_CLAUDE_DEPLOYMENT_NAME=claude-opus-4-7
AZURE_CLAUDE_SONNET_DEPLOYMENT=claude-sonnet-4-6
AZURE_CLAUDE_HAIKU_DEPLOYMENT=claude-haiku-4-6

# Azure OpenAI (gpt-5.4) — note different host from AZURE_ENDPOINT
AZURE_OPENAI_ENDPOINT=https://<resource>.cognitiveservices.azure.com/openai/v1
AZURE_GPT_DEPLOYMENT=gpt-5.4
GPT_REASONING_EFFORT=high
# Only needed with the legacy (non-/openai/v1) endpoint form:
# AZURE_OPENAI_API_VERSION=2025-04-01-preview

# Service
SERVICE_API_KEY=your-random-secret-key
PORT=3000
LOG_TOOL_CALLS=false
LOG_MESSAGES=false
    • Lưu ý: SERVICE_API_KEY để bảo vệ dịch vụ của bạn. Hãy đặt nó thành một chuỗi ký tự ngẫu nhiên.

    Cấu hình Environment Variables trên Railway

  3. Cấu hình Build Settings

    • Railway sẽ tự động detect Node.js project

  4. Deploy

    • Railway sẽ tự động deploy khi bạn push code lên GitHub
    • Hoặc click Deploy trong Railway dashboard
    • Sau khi deploy thành công, Railway sẽ cung cấp một public URL

  5. Kiểm tra Health Check

    • Truy cập: https://your-app.up.railway.app/health
    • Nếu trả về {"status":"ok"}, server đã chạy thành công

  6. Cấu hình Custom Domain (tùy chọn)

    • Vào tab Settings > Networking
    • Thêm custom domain nếu cần

    Cấu hình Custom Domain trên Railway

Lưu ý khi deploy

  • Railway tự động cung cấp PORT qua biến môi trường, nhưng bạn vẫn có thể set PORT=8080 để đảm bảo
  • SERVICE_API_KEY phải khớp chính xác với API Key bạn cấu hình trong Cursor IDE
  • Kiểm tra logs trong Railway dashboard nếu gặp lỗi

📝 License

MIT

🙏 Tham khảo

Dự án này được tham khảo từ Cursor-Azure-GPT-5 - một service cho phép Cursor sử dụng Azure GPT-5 deployments.

About

Proxy server to connect Cursor IDE with Azure Anthropic API (Claude). Transform OpenAI format to Anthropic format. Supports Claude Opus 4.5, streaming, and Railway deployment.

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages