Codex OpenAI Proxy

A proxy server that allows CLINE (Claude Code) and other OpenAI-compatible extensions to use ChatGPT Plus tokens from Codex authentication instead of requiring separate OpenAI API keys.

Overview

This proxy bridges the gap between:

• Input: Standard OpenAI Chat Completions API format (what CLINE expects)
• Output: ChatGPT Responses API format (what ChatGPT backend uses)

Features

• ✅ OpenAI API Compatibility: Accepts standard OpenAI Chat Completions requests
• ✅ ChatGPT Plus Integration: Uses your existing ChatGPT Plus tokens
• ✅ Cloudflare Bypass: Handles ChatGPT's Cloudflare protection with browser-like headers
• ✅ HTTPS Support: Works with extensions requiring secure connections (via ngrok)
• ✅ Streaming Responses: Full streaming support for real-time responses
• ✅ CLINE Compatible: Tested extensively with CLINE VS Code extension
• ✅ Array Content Support: Handles both string and array message formats from OpenAI SDK
• ✅ Universal Routing: Bulletproof request routing that bypasses complex warp conflicts

Quick Start

1. Build and Run

git clone https://github.com/Securiteru/codex-openai-proxy.git
cd codex-openai-proxy
cargo build --release
./target/release/codex-openai-proxy --port 8888 --auth-path ~/.codex/auth.json

2. Setup HTTPS Tunnel (Required for CLINE)

Most VS Code extensions require HTTPS:

# Install ngrok and create your own static domain at https://dashboard.ngrok.com/domains
# Replace 'your-static-domain' with your unique domain name
ngrok http 8888 --domain=your-static-domain.ngrok-free.app

Security Note: Always use your own unique ngrok domain. Do not share your domain publicly to prevent unauthorized access to your proxy.

3. Configure CLINE Extension

In VS Code CLINE settings:

• Base URL: https://your-static-domain.ngrok-free.app
• Model: gpt-5 (or gpt-4)
• API Key: Any value (not used, but required by extension)

4. Test Connection

# Health check
curl https://your-static-domain.ngrok-free.app/health

# Test completion
curl -X POST https://your-static-domain.ngrok-free.app/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer test-key" \
  -d '{
    "model": "gpt-5",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

How It Works

Request Flow

1. CLINE → Chat Completions format → Proxy
2. Proxy → Converts to Responses API → ChatGPT Backend
3. ChatGPT Backend → Responses API format → Proxy
4. Proxy → Converts to Chat Completions → CLINE

Format Conversion

Chat Completions Request:

{
  "model": "gpt-5",
  "messages": [
    {"role": "user", "content": "Hello!"}
  ]
}

Responses API Request:

{
  "model": "gpt-5", 
  "instructions": "You are a helpful AI assistant.",
  "input": [
    {
      "type": "message",
      "role": "user", 
      "content": [{"type": "input_text", "text": "Hello!"}]
    }
  ],
  "tools": [],
  "tool_choice": "auto",
  "store": false,
  "stream": false
}

Configuration

Command Line Options

codex-openai-proxy [OPTIONS]

Options:
  -p, --port <PORT>          Port to listen on [default: 8080]
      --auth-path <PATH>     Path to Codex auth.json [default: ~/.codex/auth.json]
  -h, --help                 Print help
  -v, --version              Print version

Authentication

The proxy automatically reads authentication from your Codex auth.json file:

{
  "access_token": "eyJ...",
  "account_id": "db1fc050-5df3-42c1-be65-9463d9d23f0b",
  "api_key": "sk-proj-..."
}

Priority: Uses access_token + account_id for ChatGPT Plus accounts, falls back to api_key for standard OpenAI accounts.

API Endpoints

Health Check

• GET /health
• Returns service status

Chat Completions

• POST /v1/chat/completions
• OpenAI-compatible chat completions endpoint
• Supports: messages, model, temperature, max_tokens, stream, tools

Troubleshooting

Common Issues

Connection Refused:

# Check if proxy is running
curl http://localhost:8080/health

Authentication Errors:

# Verify auth.json exists and has valid tokens
cat ~/.codex/auth.json | jq .

Backend Errors:

# Check proxy logs for detailed error messages
RUST_LOG=debug cargo run

Debug Mode

# Run with debug logging
RUST_LOG=debug cargo run -- --port 8080

# Test with verbose curl
curl -v -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-5", "messages": [{"role": "user", "content": "Test"}]}'

Development

Building

cargo build
cargo test
cargo clippy
cargo fmt

Adding Features

The proxy is designed to be extensible:

• New endpoints: Add routes in main.rs
• Format conversion: Modify conversion functions
• Authentication: Extend AuthData structure
• Streaming: Add SSE support for real-time responses

License

This project is part of the Codex ecosystem and follows the same licensing as the main Codex repository.