Qwen Code Proxy

A lightweight, Dockerized middleware that lets you use the official Claude Code CLI and Opencode with the Qwen3-Coder-Plus backend (via Qwen Portal).

🚀 Features

• Model Translation: Seamlessly routes to Qwen3-Coder-Plus.
• Credential Integration: Securely mounts your existing local Qwen OAuth credentials (~/.qwen/oauth_creds.json) into the container.
• Dockerized: Runs in a lightweight Python container with no dependency pollution on your host machine.

🏗️ Architecture

The Qwen Code Proxy leverages LiteLLM, an open-source AI gateway that provides a unified proxy interface for 100+ LLMs.

Core Components

1. Python Application Wrapper - Runs main.py and manages the LiteLLM proxy with retry logic and graceful shutdown
2. LiteLLM Proxy Server - Runs inside a Docker container on port 3455
3. Credential Manager - Securely accesses Qwen OAuth credentials with thread-safe caching

Request Flow

`claude` / `opencode` → Local Proxy (`3455`) → LiteLLM Translation → Qwen Portal API → Response Back to CLI

Technical Details

• Parameter Filtering: Anthropic-specific parameters like thinking and betas are automatically dropped
• Credential Caching: API keys are cached with file modification monitoring to avoid unnecessary reads
• Retry Mechanism: Automatic retries with configurable attempts and delays
• Graceful Shutdown: Signal handling for clean process termination

📋 Prerequisites

1. Docker & Docker Compose: Installed and running.
2. Claude Code CLI: Installed on your host machine (Claude Console Auth).
3. Opencode CLI: Installed on your host machine (Zen Auth).

# claude-code CLI installation
npm install -g @anthropic-ai/claude-code

# opencode installation
npm install -g opencode-ai

Qwen Credentials: You must be logged into the Qwen CLI on your machine.

🛠️ Installation & Setup

1. Clone the repository:

   git clone <repository-url>
   cd qwen-code-proxy

2. Start the Proxy: Run the Docker container in the background. This will build the image and start the LiteLLM proxy on port 3455.

   docker compose up -d

3. Verify Status: Ensure the container is running and listening:

   docker compose logs -f

   You should see 🚀 Starting Qwen Proxy on http://0.0.0.0:3455.

💻 Usage

To use the proxy, configure the Claude CLI to point to localhost instead of Anthropic's servers.

For Claude CLI

Add the following to your Claude CLI configuration file (~/.claude/settings.json):

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:3455"
  }
}

When you run claude, it will ask you to log in. Use console auth.

For Opencode CLI

Add the following to your Opencode CLI configuration file (~/.config/opencode/opencode.json):

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "litellm": {
      "name": "litellm",
      "options": {
        "baseURL": "http://127.0.0.1:3455"
      },
      "models": {
        "openai/qwen3-coder-plus": {
          "name": "openai/qwen3-coder-plus"
        }
      }
    }
  }
}

When you run opencode, it will ask you to log in. Use Zen auth.
After login, use /connect, select litellm, and enter a dummy API key (sk-xxx) to connect.

🚨 Limitations & Considerations

• Feature Parity: Some Anthropic-specific features may not be fully supported by Qwen
• Rate Limits: Subject to Qwen Portal's rate limits and usage policies
• Offline Access: Requires internet connectivity to reach Qwen Portal API

🛠️ Development

For development, you can run the proxy directly without Docker:

# Install dependencies with uv
uv sync

# Run the proxy directly with Python
uv run python main.py

# Run with custom configuration via environment variables
QWEN_LOG_LEVEL=DEBUG uv run python main.py

This approach is useful for debugging and development, bypassing the Docker container for faster iteration cycles.

❓ FAQ

• Q: How do I view the server logs?
  A: Run docker compose logs -f. This is useful for debugging connection issues or verifying that requests are hitting the Qwen API.

• Q: How do I update the project?
  A: Pull the latest changes (if any), then rebuild the container:

docker compose up -d --build

• Q: I'm getting API errors or authentication issues. How can I refresh my token?
  A: If you encounter API errors, refresh your token by navigating to the project folder and running:

cd qwen-code-proxy  # Navigate to project directory
qwen "Hello" && docker compose restart

This restarts the proxy container and refreshes the token from your credentials file. The proxy uses thread-safe caching with file modification monitoring, so it automatically picks up updated tokens when the credentials file changes.

📄 License

Only use for personal and non-commercial purposes. Do not use for any illegal or unauthorized purpose.

🙏 Acknowledgments

• LiteLLM for providing the API translation and proxy infrastructure
• Anthropic for the excellent Claude Code CLI
• Opencode for their open-source CLI tool
• Alibaba Cloud for the Qwen3-Coder-Plus model and API access