Qwen3-Coder MCP Server for Claude Code

You run an MCP server that connects the Qwen3-Coder 30B model with Claude Code through the Model Context Protocol, delivering coding assistance tools directly inside Claude Code with optimized performance on 64GB RAM systems.

How to use

Install and start the MCP server, then load Claude Code. The MCP server exposes five specialized tools you can call from your Claude Code sessions by referencing their tool names in your conversations. Use the tools to review code, explain how it works, generate new code, fix bugs, and optimize performance.

How to install

# Prerequisites: Node.js and npm must be installed
node -v
npm -v

# Clone the project or prepare the working directory
# (Assuming you already have the project files in your working directory)

# Install dependencies
npm install

# Start the optimized MCP server (runtime command shown in start script)
npm run start

Additional notes

Optimization settings and runtime recommendations are included to maximize performance on 64GB RAM systems. The server is designed to keep models loaded and responsive, with flash attention and a high-quality 8-bit KV cache.

Performance tips include keeping multiple large models loaded concurrently and tuning memory usage to avoid slowdowns during peak loads. If you notice slower responses, verify that you have sufficient RAM available and that the flash attention setting is enabled.

When you run the MCP server, you will use the stdio configuration that launches the Node.js script directly. The command is node qwen3-mcp-server.js and is intended for local execution. Ensure Claude Code can access the local MCP server endpoint if you are using a direct integration mode.

Troubleshooting

If Claude Code cannot connect to the MCP server, verify the server process is running and listening on the expected port. Check that the MCP command was started from the correct directory and that the necessary files exist.

If you encounter slow responses, confirm you have enough RAM available and that the optimization flags (flash attention, KV cache) are active. Monitor system resources to identify memory pressure or CPU bottlenecks.

Performance notes

With 64GB RAM, you can keep several large models loaded at once, handle numerous parallel requests, and use high-quality cache settings to improve responsiveness over extended runtimes.

Files and configuration

Key files include the MCP server implementation and startup scripts. The MCP server is designed to run locally and integrate with Claude Code through the defined tools.

Available tools

qwen3_code_review

Reviews code for quality, bugs, and best practices.

qwen3_code_explain

Provides detailed explanations of how code works.

qwen3_code_generate

Generates new code based on requirements.

qwen3_code_fix

Fixes bugs and issues in existing code.

qwen3_code_optimize

Optimizes code for performance, memory, or readability.