Command-Line MCP Server

A secure Model Control Protocol (MCP) server that allows AI assistants to execute terminal commands with controlled directory access and command permissions.

Overview

Command-Line MCP provides a security layer between AI assistants and your terminal. It implements a dual security model:

1. Command Permissions: Commands are categorized as read (safe), write (changes data), or system (affects system state), with different approval requirements
2. Directory Permissions: Commands can only access explicitly whitelisted directories or directories approved during a session

AI assistants interact with this server using standardized MCP tools, enabling safe terminal command execution while preventing access to sensitive files or dangerous operations. You can configure the security level from highly restrictive to more permissive based on your needs.

Key Features

Security	Usability	Integration
Directory whitelisting	Command categorization (read/write/system)	Claude Desktop compatibility
Command filtering	Persistent session permissions	Standard MCP protocol
Pattern matching	Command chaining (pipes, etc.)	Auto-approval options
Dangerous command blocking	Intuitive approval workflow	Multiple config methods

Supported Commands (out of the box)

Read Commands

• ls, pwd, cat, less, head, tail, grep, find, which, du, df, file, sort, etc.

Write Commands

• cp, mv, rm, mkdir, rmdir, touch, chmod, chown, etc.

System Commands

• ps, top, htop, who, netstat, ifconfig, ping, etc.

Security Architecture

The system implements a multi-layered security approach:

┌───────────────────────────────────────────────────────────────┐
│                   COMMAND-LINE MCP SERVER                     │
├──────────────────┬────────────────────────┬───────────────────┤
│ COMMAND SECURITY │   DIRECTORY SECURITY   │ SESSION SECURITY  │
├──────────────────┼────────────────────────┼───────────────────┤
│ ✓ Read commands  │ ✓ Directory whitelist  │ ✓ Session IDs     │
│ ✓ Write commands │ ✓ Runtime approvals    │ ✓ Persistent      │
│ ✓ System commands│ ✓ Path validation      │   permissions     │
│ ✓ Blocked list   │ ✓ Home dir expansion   │ ✓ Auto timeouts   │
│ ✓ Pattern filters│ ✓ Subdirectory check   │ ✓ Desktop mode    │
└──────────────────┴────────────────────────┴───────────────────┘

All security features can be configured from restrictive to permissive based on your threat model and convenience requirements.

Quick Start

# Install
git clone https://github.com/yourusername/cmd-line-mcp.git
cd cmd-line-mcp
python -m venv venv
source venv/bin/activate
pip install -e .
cp config.json.example config.json

# Run
cmd-line-mcp                        # With default config
cmd-line-mcp --config config.json   # With specific config

Configuration Options

The server supports four configuration methods in order of precedence:

1. Built-in default configuration (default_config.json)
2. JSON configuration file (recommended for customization)

   cmd-line-mcp --config config.json

3. Environment variables (for specific overrides)

   export CMD_LINE_MCP_SECURITY_WHITELISTED_DIRECTORIES="~,/tmp"

4. .env file (for environment-specific settings)

   cmd-line-mcp --config config.json --env .env

The default configuration is stored in default_config.json and is included with the package. You can copy this file to create your own custom configuration.

Core Configuration Settings

{
  "security": {
    "whitelisted_directories": ["/home", "/tmp", "~"],
    "auto_approve_directories_in_desktop_mode": false, 
    "require_session_id": false,
    "allow_command_separators": true
  },
  "commands": {
    "read": ["ls", "cat", "grep"], 
    "write": ["touch", "mkdir", "rm"],
    "system": ["ps", "ping"]
  }
}

Environment Variable Format

Environment variables use a predictable naming pattern:

CMD_LINE_MCP_<SECTION>_<SETTING>

Examples:

# Security settings
export CMD_LINE_MCP_SECURITY_WHITELISTED_DIRECTORIES="/projects,/var/data"
export CMD_LINE_MCP_SECURITY_AUTO_APPROVE_DIRECTORIES_IN_DESKTOP_MODE=true

# Command additions (these merge with defaults)
export CMD_LINE_MCP_COMMANDS_READ="awk,jq,wc"

Claude Desktop Integration

Setup

1. Install Claude for Desktop
2. Configure in ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "cmd-line": {
      "command": "/path/to/venv/bin/cmd-line-mcp",
      "args": ["--config", "/path/to/config.json"],
      "env": {
        "CMD_LINE_MCP_SECURITY_REQUIRE_SESSION_ID": "false",
        "CMD_LINE_MCP_SECURITY_AUTO_APPROVE_DIRECTORIES_IN_DESKTOP_MODE": "true"
      }
    }
  }
}

Recommended Claude Desktop Settings

For best experience, configure:

• require_session_id: false - Essential to prevent approval loops
• auto_approve_directories_in_desktop_mode: true - Optional for convenient access
• Include common directories in your whitelist

After configuration, restart Claude for Desktop.

AI Assistant Tools

The server provides these MCP tools for AI assistants:

Tool	Purpose	Needs Approval
execute_command	Run any command type	Yes, for write/system commands
execute_read_command	Run read-only commands	Directory approval only
approve_directory	Grant access to a directory	N/A - it's an approval tool
approve_command_type	Grant permission for command category	N/A - it's an approval tool
list_directories	Show authorized directories	No
list_available_commands	Show command categories	No
get_command_help	Get command usage guidance	No
get_configuration	View current settings	No

Tool Examples

Directory Management

# Check available directories
dirs = await list_directories(session_id="session123")
whitelisted = dirs["whitelisted_directories"]
approved = dirs["session_approved_directories"]

# Request permission for a directory
if "/projects/my-data" not in whitelisted and "/projects/my-data" not in approved:
    result = await approve_directory(
        directory="/projects/my-data", 
        session_id="session123"
    )

Command Execution

# Read commands (read permissions enforced)
result = await execute_read_command("ls -la ~/Documents")

# Any command type (may require command type approval)
result = await execute_command(
    command="mkdir -p ~/Projects/new-folder", 
    session_id="session123"
)

Get Configuration

# Check current settings
config = await get_configuration()
whitelist = config["directory_whitelisting"]["whitelisted_directories"]

Directory Security System

The server restricts command execution to specific directories, preventing access to sensitive files.

Directory Security Modes

The system supports three security modes:

Mode	Description	Best For	Configuration
Strict	Only whitelisted directories allowed	Maximum security	auto_approve_directories_in_desktop_mode: false
Approval	Non-whitelisted directories require explicit approval	Interactive use	Default behavior for standard clients
Auto-approve	Auto-approves directories for Claude Desktop	Convenience	auto_approve_directories_in_desktop_mode: true

Whitelisted Directory Configuration

"security": {
  "whitelisted_directories": [
    "/home",                  // System directories
    "/tmp",
    "~",                      // User's home
    "~/Documents"             // Common user directories
  ],
  "auto_approve_directories_in_desktop_mode": false  // Set to true for convenience
}

Directory Approval Flow

1. Command is requested in a directory
2. System checks:
   • Is the directory in the global whitelist? → Allow
   • Has directory been approved in this session? → Allow
   • Neither? → Request approval
3. After approval, directory remains approved for the entire session

Path Format Support

• Absolute paths: /home/user/documents
• Home directory: ~ (expands to user's home)
• User subdirectories: ~/Downloads

Claude Desktop Integration

The server maintains a persistent session for Claude Desktop, ensuring directory approvals persist between requests and preventing approval loops.

Command Customization

The system uses command categorization to control access:

Category	Description	Example Commands	Requires Approval
Read	Safe operations	ls, cat, find	No
Write	Data modification	mkdir, rm, touch	Yes
System	System operations	ps, ping, ifconfig	Yes
Blocked	Dangerous commands	sudo, bash, eval	Always denied

Customization Methods

// In config.json
{
  "commands": {
    "read": ["ls", "cat", "grep", "awk", "jq"],
    "write": ["mkdir", "touch", "rm"],
    "system": ["ping", "ifconfig", "kubectl"],
    "blocked": ["sudo", "bash", "eval"]
  }
}

Environment Variable Method:

# Add to existing lists, not replace (comma-separated)
export CMD_LINE_MCP_COMMANDS_READ="awk,jq"
export CMD_LINE_MCP_COMMANDS_BLOCKED="npm,pip"

The MCP server merges these additions with existing commands, letting you extend functionality without recreating complete command lists.

Command Chaining

The server supports three command chaining methods:

Method	Symbol	Example	Config Setting
Pipes	|	ls | grep txt	allow_command_separators: true
Sequence	;	mkdir dir; cd dir	allow_command_separators: true
Background	&	find . -name "*.log" &	allow_command_separators: true

All commands in a chain must be from the supported command list. Security checks apply to the entire chain.

Quick Configuration:

"security": {
  "allow_command_separators": true  // Set to false to disable all chaining
}

To disable specific separators, add them to the dangerous_patterns list.

License

MIT