factory-cursor-bridge

A unified BYOK proxy that wires any OpenAI-compatible model into Cursor IDE via the "Override OpenAI Base URL" setting. Supports multiple providers, protocol translation (OpenAI ↔ Anthropic), and automatic model prefixing.

How it works

Cursor IDE
    │
    ├─ Override OpenAI Base URL ──► https://your-tunnel-url.trycloudflare.com
    │   (OpenAI API Key = bearer token)
    │
    ▼
factory-cursor-bridge (:8316)
    │
    ├─ strips custom prefix (e.g. fx-) from model name
    ├─ looks up base URL & API key from config file
    ├─ translates protocol (OpenAI ↔ Anthropic) if needed
    │
    ▼
Your actual API providers

Quick Start

1. Configure your models

Create a config.json file with your models:

{
  "custom_models": [
    {
      "model": "my-gpt-4",
      "model_display_name": "My GPT-4",
      "base_url": "https://api.openai.com/v1",
      "api_key": "sk-...",
      "provider": "openai"
    },
    {
      "model": "my-claude",
      "model_display_name": "My Claude",
      "base_url": "https://api.anthropic.com/v1",
      "api_key": "sk-ant-...",
      "provider": "anthropic"
    },
    {
      "model": "my-local-model",
      "model_display_name": "Local Model",
      "base_url": "http://localhost:11434/v1",
      "api_key": "local",
      "provider": "openai"
    }
  ]
}

2. Start the proxy

node factory-cursor-bridge.mjs --config ./config.json --port 8316

Or set the environment variable:

export FACTORY_CONFIG=/path/to/your/config.json
node factory-cursor-bridge.mjs

3. Expose via tunnel

Cursor blocks 127.0.0.1 (SSRF protection). You need a public tunnel:

cloudflared tunnel --url http://127.0.0.1:8316

Use the .trycloudflare.com URL it gives you.

4. Configure Cursor

1. Settings → Models
2. Set OpenAI API Key to the bearer token from your config
3. Enable Override OpenAI Base URL → your tunnel URL
4. Click + Add Custom Model and add your model names with the fx- prefix (e.g., fx-my-gpt-4, fx-my-claude)

Config File Format

{
  "custom_models": [
    {
      "model": "model-id-in-upstream",
      "model_display_name": "Display Name in Cursor",
      "base_url": "https://api.provider.com/v1",
      "api_key": "your-api-key",
      "provider": "openai | anthropic | generic-chat-completion-api",
      "supports_images": true,
      "extra_headers": { "X-Custom-Header": "value" },
      "extra_args": { "reasoning_effort": "high" }
    }
  ]
}

Provider types

Provider	Protocol	Notes
openai	OpenAI	Standard chat completions
anthropic	Anthropic	Auto-converts to Anthropic Messages API
generic-chat-completion-api	OpenAI	Generic OpenAI-compatible backend

Model naming

Each model in your config gets a fx- prefix in Cursor:

Config model	Cursor model name
gpt-4o	fx-gpt-4o
claude-3-5-sonnet	fx-claude-3-5-sonnet
qwen-72b	fx-qwen-72b

Full example with multiple providers

{
  "custom_models": [
    {
      "model": "gpt-5.4",
      "model_display_name": "GPT-5.4",
      "base_url": "http://localhost:8318/v1",
      "api_key": "local",
      "provider": "openai"
    },
    {
      "model": "claude-opus-4-6",
      "model_display_name": "Claude Opus",
      "base_url": "http://localhost:8318/v1",
      "api_key": "local",
      "provider": "anthropic"
    },
    {
      "model": "llama-3.1-70b",
      "model_display_name": "Llama 70B",
      "base_url": "https://api.groq.com/openai/v1",
      "api_key": "gsk_...",
      "provider": "openai"
    },
    {
      "model": "gemini-1.5-pro",
      "model_display_name": "Gemini Pro",
      "base_url": "https://generativelanguage.googleapis.com/v1beta",
      "api_key": "AI...",
      "provider": "generic-chat-completion-api"
    },
    {
      "model": "deepseek-coder",
      "model_display_name": "DeepSeek Coder",
      "base_url": "http://localhost:11434/v1",
      "api_key": "local",
      "provider": "openai"
    }
  ]
}

CLI Options

node factory-cursor-bridge.mjs [options]

Options:
  --config <path>   Path to config.json (default: ~/.factory/config.json)
  --port <port>     Port to listen on (default: 8316)
  --host <host>     Host to bind to (default: 127.0.0.1)
  --token <token>   Bearer token for auth (default: from config)

Endpoints

Method	Path	Description
GET	/health	Health check
GET	/v1/models	List available models
POST	/v1/chat/completions	Chat completions

Troubleshooting

"Provider was unable to process your request" error → Cursor blocks 127.0.0.1. You must use a public tunnel URL.

"Model not found" error → Make sure you're using the fx- prefixed name in Cursor.

Anthropic models not working → The proxy auto-converts OpenAI format to Anthropic format for anthropic provider type.