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.
- Large Context Windows: Leverage Qwen's massive token capacity for analyzing large files and entire codebases
- File Analysis: Use
@filenameor@directorysyntax 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
- Node.js v16 or higher
- Qwen CLI installed and configured (qwen-code)
Use Claude Code's built-in MCP installer:
claude mcp add qwen-cli -- npx -y @jaggerxtrm/qwen-mcp-toolThis single command configures everything automatically!
Install via npm:
npm install -g @jaggerxtrm/qwen-mcp-toolThen add to Claude Code MCP settings (~/.config/claude/mcp_settings.json):
{
"mcpServers": {
"qwen-cli": {
"command": "qwen-mcp-tool"
}
}
}Manually configure to use npx without installing:
{
"mcpServers": {
"qwen-cli": {
"command": "npx",
"args": ["-y", "@jaggerxtrm/qwen-mcp-tool"]
}
}
}- Clone and install dependencies:
git clone <repo-url>
cd qwen-mcp-tool
npm install- Build the project:
npm run build- Link locally:
npm linkThe main tool for interacting with Qwen AI.
Parameters:
prompt(required): Your question or instruction- Use
@filenameto include a file's contents - Use
@directoryto include all files in a directory
- Use
model(optional): Model to use (qwen3-coder-plus, qwen3-coder-turbo, etc.)sandbox(optional): Enable sandbox mode for safe code executionapprovalMode(optional): Control tool execution approvalplan: Analyze tool calls without executingdefault: Prompt for approval (default behavior)auto-edit: Auto-approve file editsyolo: Auto-approve all tool calls
yolo(optional): Shortcut for approvalMode='yolo'allFiles(optional): Include all files in current directory as contextdebug(optional): Enable debug mode
Examples:
// Analyze a specific file
{
"prompt": "@src/main.ts Explain what this code does"
}
// Analyze entire codebase
{
"prompt": "@src/ Summarize the architecture of this codebase"
}
// Use specific model with sandbox
{
"prompt": "Run the test suite and fix any failures",
"model": "qwen3-coder-plus",
"sandbox": true,
"approvalMode": "auto-edit"
}Simple echo test to verify the connection.
Parameters:
prompt(optional): Message to echo (defaults to "Pong!")
Display Qwen CLI help information.
Parameters: None
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.
Once installed as an MCP server, you can use it within Claude Code:
Ask Qwen to analyze the authentication system in @src/auth/
Claude will automatically use the ask-qwen tool with the appropriate parameters.
qwen-mcp-tool/
├── src/
│ ├── index.ts # MCP server entry point
│ ├── constants.ts # Configuration and constants
│ ├── tools/
│ │ ├── registry.ts # Tool registration system
│ │ ├── ask-qwen.tool.ts # Main Qwen interaction tool
│ │ ├── simple-tools.ts # Utility tools (ping, help)
│ │ └── index.ts # Tool exports
│ └── utils/
│ ├── commandExecutor.ts # Command execution utility
│ ├── qwenExecutor.ts # Qwen CLI wrapper
│ └── logger.ts # Logging utility
├── package.json
├── tsconfig.json
└── README.md
- The MCP server listens for tool calls via stdio transport
- When a tool is called, the server validates the arguments using Zod schemas
- For
ask-qwen, the prompt is passed to the Qwen CLI with appropriate flags - File references (
@filename) are handled by Qwen's built-in file processing - Output is captured and returned to the MCP client
- If quota limits are hit, the server automatically falls back to the turbo model
This tool is inspired by gemini-mcp-tool but adapted for Qwen CLI:
| Feature | Gemini MCP | Qwen MCP |
|---|---|---|
| File references | ✅ | ✅ (more advanced) |
| Sandbox mode | ✅ | ✅ |
| Multiple models | ✅ | ✅ |
| Approval modes | ❌ | ✅ |
| Directory traversal | Basic | Advanced (git-aware) |
| Multimodal support | Limited | Images, PDFs, audio, video |
Make sure the Qwen CLI is installed and available in your PATH:
npm install -g @qwen/cli
# or follow instructions at https://github.com/QwenLM/qwen-codeFor very large files or codebases, the analysis may take longer than the default 10-minute timeout. Consider:
- Using
.qwenignoreto exclude unnecessary files - Breaking down large queries into smaller chunks
- Using
approvalMode: "plan"to analyze without executing
Check that your arguments match the tool schema. Use the Help tool to see available options.
MIT
Contributions are welcome! Please feel free to submit issues or pull requests.
Inspired by gemini-mcp-tool by jamubc. Built for use with Qwen Code.