Kommand

An MCP server for spawning isolated Qwen sub-agents with secure project isolation and dynamic tool management.

Features

• Dynamic Agent Discovery: Automatically discovers and loads agent configurations
• Project Isolation: Each agent operates in isolated environments with separate workspaces
• Security-First Design: Comprehensive file safety checks and environment variable protection
• Conversation Management: Persistent history and context sharing between agents
• Dynamic Tool Generation: Auto-creates MCP tools for each discovered agent type
• Flexible Configuration: Customizable timeouts, cleanup policies, and agent permissions

Installation

1. Clone the repository:

   git clone https://github.com/greenmutton/kommand.git
   cd kommand

2. Install dependencies:

   pip install fastmcp==2.11.0

3. Set up environment variables:

   export KOMMAND_BASE_DIR="$HOME/base/kommand"
   export KOMMAND_VENV="$HOME/base/venvs/kommand/bin/activate"

Notes

Currently this project has only been tested with qwen-code (a fork of gemini-cli). However it can easily be extended to work with any CLI agent that supports non-interactive task-passing via stdin. Explicit support for other CLI agents is planned in a later release.

A more advanced organization system for handling a large number of specialist agents is also in the plans.

Configuration

Create agent directories under $KOMMAND_BASE_DIR/agents/ with:

• specialization.json - Agent configuration
• system_prompt.txt - Agent's system prompt
• .env - Environment variables, such as OPENAI_API_KEY, OPENAI_BASE_URL, OPENAI_MODEL

Example specialization.json:

{
  "mcp_servers": ["filesystem", "git"],
  "sub_agents": ["researcher", "analyst"],
  "timeout_sec": 30000,
  "cleanup_after": true,
  "max_workspace_files": 50
}

MCP Usage

"mcpServers": {
  "mcp-kommander": {
    "command": "/path/to/venvs/kommander-venv/bin/activate",
    "args": [
      "/path/to/mcp_kommander.py"
    ],
    "env": {
      "KOMMAND_BASE_DIR": "/path/to/base/dir",
      "KOMMAND_VENV": "/path/to/venvs/kommander-venv/bin/activate"
    }
  }
}

Agent Structure

Each agent has:

• workspace/ - Temporary working files
• output/ - Generated artifacts
• logs/ - Execution logs
• conversation_history.json - Persistent conversation history

Security

• File safety checks with extension and content validation
• Sensitive environment variable protection
• Path traversal prevention
• Execution permission restrictions
• Workspace file count and size limits

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

Contributors

GLM-4.5, K2, @greenmutton

License

MIT License - see LICENSE file for details.