dynamic-shell-server

A Model Context Protocol (MCP) server that enables secure execution of shell commands with a dynamic approval system. This server allows running arbitrary commands while maintaining security through user approval and audit logging.

• 🔐 Dynamic command approval system
• 📝 Persistent storage of approved commands
• 📊 Comprehensive audit logging
• ⏱️ Command timeout protection
• 🔄 Command revocation capability

1. Clone this repository:

git clone <repository-url>
cd dynamic-shell-server

2. Create a virtual environment and activate it:

python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Run the server directly:

python dynamic_shell_server.py

1. Open your Claude Desktop configuration:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json

2. Add the server configuration:

{
    "mcpServers": {
        "shell": {
            "command": "/absolute/path/to/.venv/bin/python",
            "args": ["/absolute/path/to/dynamic_shell_server.py"]
        }
    }
}

3. Restart Claude Desktop

When running a command for the first time, you'll be prompted with:

Command Approval Required

Command: <command>
Arguments: <args>

This command has not been previously approved. Do you want to:
1. Allow this command once
2. Allow this command and remember for future use
3. Deny execution

Please choose an option (1-3):

1. execute_command: Execute a shell command

   • Parameters:
     • command: The command to execute
     • args: Optional list of command arguments

2. revoke_command_approval: Revoke approval for a previously approved command

   • Parameters:
     • command: The command to revoke approval for

1. commands://approved: Lists all approved commands with their approval dates

The server stores its data in ~/.config/mcp-shell-server/:

• approved_commands.json: List of approved commands and their approval dates
• audit.log: Detailed execution history of all commands

• User approval required for first-time command execution
• Persistent storage of approved commands
• Comprehensive audit logging
• 5-minute command timeout
• No shell execution (prevents injection attacks)
• Command revocation capability

Through Claude Desktop:

Human: Run 'npm install' in my project directory
Assistant: I'll help you run that command. Since this is the first time running 'npm install', you'll need to approve it.
[Command approval prompt appears]