Qwen MCP Tool

Model Context Protocol server for Qwen CLI integration. This tool enables AI assistants like Claude to leverage Qwen's powerful code analysis and large context window capabilities through the MCP protocol.

Features

• Large Context Windows: Leverage Qwen's massive token capacity for analyzing large files and entire codebases

• File Analysis: Use @filename or @directory syntax to include file contents in your queries

• Sandbox Mode: Safely execute code and run tests in isolated environments

• Multiple Models: Support for various Qwen models (qwen3-coder-plus, qwen3-coder-turbo, etc.)

• Flexible Approval Modes: Control tool execution with plan/default/auto-edit/yolo modes

• MCP Protocol: Seamless integration with MCP-compatible AI assistants

Prerequisites

• Node.js v16 or higher

• Qwen CLI installed and configured (qwen-code)

Installation

Quick Setup (Easiest - Recommended)

Use Claude Code's built-in MCP installer:

This single command configures everything automatically!

Via Global Install

Install via npm:

Then add to Claude Code MCP settings (~/.config/claude/mcp_settings.json):

Via npx (Manual Configuration)

Manually configure to use npx without installing:

From Source (Development)

1. Clone and install dependencies:

2. Build the project:

3. Link locally:

Available Tools

ask-qwen

The main tool for interacting with Qwen AI.

Parameters:

• prompt (required): Your question or instruction

  • Use @filename to include a file's contents

  • Use @directory to include all files in a directory

• model (optional): Model to use (qwen3-coder-plus, qwen3-coder-turbo, etc.)

• sandbox (optional): Enable sandbox mode for safe code execution

• approvalMode (optional): Control tool execution approval

  • plan: Analyze tool calls without executing

  • default: Prompt for approval (default behavior)

  • auto-edit: Auto-approve file edits

  • yolo: Auto-approve all tool calls

• yolo (optional): Shortcut for approvalMode='yolo'

• allFiles (optional): Include all files in current directory as context

• debug (optional): Enable debug mode

Examples:

ping

Simple echo test to verify the connection.

Parameters:

• prompt (optional): Message to echo (defaults to "Pong!")

Help

Display Qwen CLI help information.

Parameters: None

Configuration

The tool uses the following default models:

• Primary: qwen3-coder-plus

• Fallback: qwen3-coder-turbo (used if primary hits quota limits)

You can override these by specifying the model parameter in your requests.

Usage with Claude Code

Once installed as an MCP server, you can use it within Claude Code:

Claude will automatically use the ask-qwen tool with the appropriate parameters.

Project Structure

How It Works

1. The MCP server listens for tool calls via stdio transport

2. When a tool is called, the server validates the arguments using Zod schemas

3. For ask-qwen, the prompt is passed to the Qwen CLI with appropriate flags

4. File references (@filename) are handled by Qwen's built-in file processing

5. Output is captured and returned to the MCP client

6. If quota limits are hit, the server automatically falls back to the turbo model

Comparison with Gemini MCP Tool

This tool is inspired by gemini-mcp-tool but adapted for Qwen CLI:

Troubleshooting

"Qwen CLI not found"

Make sure the Qwen CLI is installed and available in your PATH:

"Command timed out"

For very large files or codebases, the analysis may take longer than the default 10-minute timeout. Consider:

• Using .qwenignore to exclude unnecessary files

• Breaking down large queries into smaller chunks

• Using approvalMode: "plan" to analyze without executing

"Invalid tool arguments"

Check that your arguments match the tool schema. Use the Help tool to see available options.

License

MIT

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Credits

Inspired by gemini-mcp-tool by jamubc. Built for use with Qwen Code.