sudo-mcp

MCP (Model Context Protocol) server that allows AI models to execute commands with elevated privileges via sudo. On Linux, privilege escalation uses polkit/pkexec with your desktop authentication agent. On macOS, it uses sudo -A with a native password dialogue via osascript.

Caution

SECURITY WARNING

As G.K. Chesterton said: "Don't ever take a fence down until you know the reason it was put up".

This is an inherently dangerous tool. By design, it allows an AI model to execute commands with root privileges on your system. While basic safeguards can be configured (command blocklist, audit logging), NO SECURITY MEASURE IS PERFECT.

USE AT YOUR OWN RISK. This tool could potentially:

Destroy data through file system operations
Compromise system security by modifying critical configurations
Modify or delete critical system files including kernel and boot files
Execute malicious commands if the blocklist is bypassed or disabled
Escalate privileges beyond what you intended
Expose sensitive information through command output

The --no-blocklist flag removes ALL command validation. Using this option gives the AI unrestricted root access to your system.

You have been warned.

Despite this, this tool is still safer than OpenClaw, at least you're asked for your password to do dumb stuff.

Overview

sudo-mcp is a C# MCP server that integrates with Claude Desktop (or any MCP client) to enable execution of privileged commands. On Linux it uses polkit/pkexec for authentication; on macOS it uses sudo -A with a native osascript password dialogue. Both platforms support configurable command validation.

Key Features:

🔐 Platform-Native Privilege Escalation - polkit/pkexec on Linux, sudo with native macOS password dialogue
✅ Configurable Command Validation - Optional blocklist to prevent dangerous operations (Linux and macOS patterns)
📝 Comprehensive Audit Logging - Every command attempt logged with full details
⚙️ Runtime Configuration - Command-line arguments for blocklist, timeouts, and logging
🚀 .NET 10 - Built on the latest .NET platform with C# 12
🍎 Cross-Platform - Supports Linux (x64/ARM64) and macOS (Intel/Apple Silicon)

Quick Start (Claude Code)

macOS: Download from releases, extract, run ./install.sh, then:

claude mcp add sudo-mcp /usr/local/bin/sudo-mcp           # Configure Claude Code

Arch Linux:

yay -S sudo-mcp                            # Install from AUR
claude mcp add sudo-mcp /usr/bin/sudo-mcp  # Configure Claude Code

Fedora/RHEL/CentOS:

sudo dnf copr enable hughesjs/sudo-mcp     # Enable COPR repository
sudo dnf install sudo-mcp                  # Install from COPR
claude mcp add sudo-mcp /usr/bin/sudo-mcp  # Configure Claude Code

Other distributions: Download from releases, extract, run ./install.sh, then use claude mcp add.

Restart Claude Code and approve authentication prompts when commands execute. Read the security warning above before use.

Prerequisites

(If Building) .NET 10 SDK - Install from dotnet.microsoft.com
Linux or macOS - Supported platforms
Linux: Polkit authentication agent - Required for graphical authentication (typically included in desktop environments)
macOS: sudo (included by default) - Uses native osascript password dialogue for authentication
Claude Desktop - Or any MCP-compatible client

Installation

Binary Release (Linux and macOS)

Step 1: Visit the Releases Page to download the latest version for your architecture.

Step 2: Extract and install:

x86_64 (Intel/AMD):

# Replace VERSION with the version you downloaded (e.g., 0.1.0)
tar -xzf sudo-mcp-x64-vVERSION.tar.gz
cd sudo-mcp-x64-vVERSION
chmod +x install.sh
./install.sh

ARM64 (aarch64):

# Replace VERSION with the version you downloaded (e.g., 0.1.0)
tar -xzf sudo-mcp-arm64-vVERSION.tar.gz
cd sudo-mcp-arm64-vVERSION
chmod +x install.sh
./install.sh

Arch Linux (AUR)

From AUR (Recommended):

yay -S sudo-mcp

Manual PKGBUILD:

If you prefer to build manually, visit the Releases Page, download the PKGBUILD, and run:

makepkg -si

Fedora/RHEL/CentOS (COPR)

From COPR (Recommended):

# Enable the COPR repository
sudo dnf copr enable hughesjs/sudo-mcp

# Install sudo-mcp
sudo dnf install sudo-mcp

Supported distributions:

Fedora 40, 41, 42, Rawhide (x86_64, aarch64)
EPEL 9 (x86_64, aarch64)

Manual specfile:

If you prefer to build manually, visit the Releases Page, download the sudo-mcp.spec file, and use rpmbuild or mock.

From Source (Development)

git clone https://github.com/hughesjs/sudo-mcp.git
cd sudo-mcp
chmod +x scripts/install.sh
./scripts/install.sh

This builds from source and installs to /usr/bin/sudo-mcp with default configuration.

Manual Build and Installation

# Clone and build
git clone https://github.com/hughesjs/sudo-mcp.git
cd sudo-mcp

# Build self-contained binary
dotnet publish src/SudoMcp/SudoMcp.csproj \
    -c Release \
    -r linux-x64 \
    --self-contained \
    -o ./publish \
    /p:PublishSingleFile=true

# Install system-wide
sudo cp publish/SudoMcp /usr/bin/sudo-mcp
sudo chmod +x /usr/bin/sudo-mcp

# Set up log directory
sudo mkdir -p /var/log/sudo-mcp
sudo chown $USER:$USER /var/log/sudo-mcp

Verify Installation

sudo-mcp --help

Uninstall

chmod +x scripts/uninstall.sh
./scripts/uninstall.sh

Command-Line Options

sudo-mcp supports the following command-line arguments for runtime configuration:

Option	Short	Description	Default
`--blocklist-file <path>`	`-b`	Path to custom blocklist JSON file	Embedded default
`--no-blocklist`	-	DANGEROUS: Disable all command validation	`false`
`--audit-log <path>`	`-a`	Path to audit log file	Linux: `/var/log/sudo-mcp/audit.log`, macOS: `~/Library/Logs/sudo-mcp/audit.log`
`--timeout <seconds>`	`-t`	Command execution timeout in seconds	`15`

Examples

Default configuration (with blocklist):

sudo-mcp

Custom blocklist:

sudo-mcp --blocklist-file /path/to/custom-blocklist.json

No blocklist (MAXIMUM DANGER):

sudo-mcp --no-blocklist

Custom timeout and audit log:

sudo-mcp --timeout 60 --audit-log /home/user/.sudo-mcp/audit.log

Display help:

sudo-mcp --help

Configuration

Blocklist Configuration

sudo-mcp includes an embedded default blocklist that prevents execution of dangerous commands using three strategies:

Exact Matches - Specific dangerous commands (e.g., rm -rf /)
Regex Patterns - Pattern-based blocking for classes of operations
Blocked Binaries - Specific executables regardless of arguments

Example blocklist:

{
  "BlockedCommands": {
    "ExactMatches": [
      "rm -rf /",
      "mkfs",
      "dd"
    ],
    "RegexPatterns": [
      "^rm\\s+(-rf?|--recursive)\\s+/\\s*$",
      "^dd\\s+if=.+\\s+of=/dev/(sd[a-z]|nvme[0-9]n[0-9]).*$",
      "^mkfs\\..*"
    ],
    "BlockedBinaries": [
      "mkfs.ext4",
      "shred",
      "cryptsetup"
    ]
  }
}

Example blocklist files are provided in the examples/ directory:

blocklist-default.json - Exact copy of embedded default (reference implementation)
blocklist-permissive.json - Relaxed rules for development environments
blocklist-strict.json - Enhanced security for production-adjacent environments
blocklist-minimal.json - Bare minimum for testing in disposable VMs

See examples/blocklist-README.md for detailed documentation.

Using an example blocklist:

sudo-mcp --blocklist-file examples/blocklist-permissive.json

Customising the blocklist:

Create your own JSON file and pass it via --blocklist-file:

sudo-mcp --blocklist-file /etc/sudo-mcp/my-blocklist.json

Disabling the blocklist:

⚠️ WARNING: Only use --no-blocklist in isolated/test environments where you fully accept the risk.

sudo-mcp --no-blocklist

Audit Logging

All command execution attempts (both allowed and denied) are logged in JSON format:

{
  "Timestamp": "2026-01-11T18:45:00Z",
  "EventType": "CommandExecuted",
  "Command": "systemctl restart nginx",
  "User": "james",
  "ExitCode": 0,
  "Success": true
}

Custom log location:

sudo-mcp --audit-log /home/user/.sudo-mcp/audit.log

Ensure the log directory exists and is writable:

sudo mkdir -p /var/log/sudo-mcp
sudo chown $USER:$USER /var/log/sudo-mcp

MCP Client Integration

After installation, configure your MCP client to use sudo-mcp.

Claude Desktop

Configuration file location:

Linux: ~/.config/Claude/claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Basic configuration:

{
  "mcpServers": {
    "sudo-mcp": {
      "command": "/usr/bin/sudo-mcp"
    }
  }
}

On macOS, use /usr/local/bin/sudo-mcp (or the Homebrew prefix path) instead of /usr/bin/sudo-mcp.

Custom configuration:

{
  "mcpServers": {
    "sudo-mcp": {
      "command": "/usr/bin/sudo-mcp",
      "args": [
        "--blocklist-file",
        "/home/YOUR_USERNAME/.config/sudo-mcp/blocklist.json",
        "--timeout",
        "30"
      ]
    }
  }
}

Claude Code

Using the CLI (Recommended):

After installing the binary, simply run:

# Linux
claude mcp add sudo-mcp /usr/bin/sudo-mcp

# macOS
claude mcp add sudo-mcp /usr/local/bin/sudo-mcp

This automatically configures sudo-mcp in your Claude Code settings.

With custom arguments:

claude mcp add sudo-mcp /usr/bin/sudo-mcp -- --timeout 30 --blocklist-file /path/to/custom-blocklist.json

Manual configuration:

Alternatively, you can manually edit the configuration file:

Linux: ~/.config/claude/config.json
macOS: ~/Library/Application Support/Claude/config.json

{
  "mcpServers": {
    "sudo-mcp": {
      "command": "/usr/bin/sudo-mcp"
    }
  }
}

Cursor

Add to your Cursor MCP settings or project-specific .cursor/mcp.json:

{
  "mcpServers": {
    "sudo-mcp": {
      "command": "/usr/bin/sudo-mcp"
    }
  }
}

See examples/ directory for more configuration examples.

Usage Examples

Once configured with Claude Desktop, you can ask Claude to execute commands:

Safe command:

User: "Check if nginx is running"
Claude: [Uses sudo-mcp to execute: systemctl status nginx]

Command requiring authentication:

User: "Restart the nginx service"
Claude: [Uses sudo-mcp to execute: systemctl restart nginx]
[Polkit authentication dialog appears]
[User approves the action]

Blocked command (with blocklist enabled):

User: "Format /dev/sda"
Claude: [sudo-mcp blocks the command]
"This command is blocked for safety: matches dangerous pattern"

How It Works

Claude requests command execution via MCP protocol over stdio
sudo-mcp receives the request and validates the command
CommandValidator checks the command against the blocklist (if enabled)
If allowed, the platform executor runs the command with elevated privileges:
- Linux: pkexec sudo <command> - polkit authentication dialogue appears
- macOS: sudo -A <command> - native macOS password dialogue appears via osascript
User approves or denies the privilege escalation
Command executes (if approved) and output is captured
AuditLogger records the execution attempt with full details
Results returned to Claude with stdout/stderr/exit code

Architecture

Claude Desktop (MCP Client)
    ↓ JSON-RPC over stdio
MCP Server (stdio transport)
    ↓
SudoExecutionTool
    ↓
CommandValidator → [validate against blocklist]
    ↓
IPrivilegedExecutor
    ├─ PkexecExecutor (Linux) → [spawn pkexec → sudo process]
    └─ SudoExecutor (macOS)   → [spawn sudo -A with osascript askpass]
    ↓
AuditLogger → [log to audit file]

Security Considerations

See SECURITY.md for comprehensive security documentation.

Key Limitations

Blocklist bypass: Regex patterns can potentially be circumvented
LLM context: The AI may not fully understand command implications
No rollback: Executed commands cannot be undone
Log tampering: Audit logs can be deleted with sudo access
Linux GUI required: pkexec requires a polkit authentication agent (graphical session)
macOS credential caching: sudo caches credentials for a timeout period, meaning subsequent commands may not prompt

Recommended Practices

Run in isolated environment: Use a dedicated VM or container
Review audit logs regularly: Check /var/log/sudo-mcp/audit.log
Start with blocklist: Only use --no-blocklist when absolutely necessary
Test in safe environment: Verify behaviour before production use
Monitor system changes: Use file integrity monitoring (e.g., AIDE, Tripwire)

Troubleshooting

pkexec authentication fails

Problem: Authorization failed (exit code 127)

Solution: Ensure your user is in the appropriate group (e.g., sudo, wheel) and polkit policies allow the action:

groups $USER
pkexec whoami  # Test pkexec authentication

Authentication dialog doesn't appear

Problem: No polkit dialog shown

Solution: Ensure a polkit authentication agent is running:

# Check for polkit agent
ps aux | grep polkit

# Install polkit-gnome (GNOME) or polkit-kde (KDE)
sudo pacman -S polkit-gnome  # Arch Linux
sudo apt install policykit-1-gnome  # Debian/Ubuntu
sudo dnf install polkit-gnome  # Fedora/RHEL

macOS password dialogue doesn't appear

Problem: No authentication dialogue shown on macOS

Solution: Ensure osascript is available (included by default on macOS). If running in a headless environment (SSH session, CI), the osascript dialogue cannot be displayed. sudo-mcp on macOS requires a graphical session.

Audit log not created

Problem: Audit log file not being written

Solution: Ensure the directory exists and is writable:

sudo mkdir -p /var/log/sudo-mcp
sudo chown $USER:$USER /var/log/sudo-mcp

Or use a custom location:

dotnet run -- --audit-log /home/$USER/.sudo-mcp/audit.log

Command timeout

Problem: Long-running commands are terminated

Solution: Increase the timeout value:

dotnet run -- --timeout 1800  # 30 minutes

Development

Building

dotnet build

Running Tests

All Tests (requires Docker):

dotnet test

Unit Tests Only (fast, no Docker):

dotnet test --filter "Category!=Integration"

Integration Tests Only (requires Docker):

dotnet test --filter "Category=Integration"

Tests use TestContainers.NET to automatically manage Docker containers. See src/SudoMcp.Tests/Integration/README.md for details.

Project Structure

sudo-mcp/
├── src/
│   ├── SudoMcp/              # Main application
│   │   ├── Tools/            # MCP tool implementations
│   │   ├── Services/         # Core services (validator, executor, logger)
│   │   ├── Models/           # Data models
│   │   └── Configuration/    # Config files (blocklist, appsettings)
│   └── SudoMcp.Tests/        # Unit and integration tests
├── examples/                 # Example configurations
├── README.md
├── SECURITY.md
└── CLAUDE_DESKTOP_SETUP.md

Contributing

Contributions are welcome! Please:

Read SECURITY.md to understand security implications
Follow existing code style (C# conventions)
Add tests for new functionality
Update documentation as needed

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

Built with Model Context Protocol
Uses System.CommandLine for argument parsing
Powered by .NET 10

Name		Name	Last commit message	Last commit date
Latest commit History 22 Commits
.github/workflows		.github/workflows
examples		examples
scripts		scripts
src		src
.dockerignore		.dockerignore
.gitattributes		.gitattributes
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
sudo-mcp.slnx		sudo-mcp.slnx

Folders and files

Latest commit

History

Repository files navigation

sudo-mcp

SECURITY WARNING

Overview

Quick Start (Claude Code)

Prerequisites

Installation

Binary Release (Linux and macOS)

Arch Linux (AUR)

Fedora/RHEL/CentOS (COPR)

From Source (Development)

Manual Build and Installation

Verify Installation

Uninstall

Command-Line Options

Examples

Configuration

Blocklist Configuration

Audit Logging

MCP Client Integration

Claude Desktop

Claude Code

Cursor

Usage Examples

How It Works

Architecture

Security Considerations

Key Limitations

Recommended Practices

Troubleshooting

pkexec authentication fails

Authentication dialog doesn't appear

macOS password dialogue doesn't appear

Audit log not created

Command timeout

Development

Building

Running Tests

Project Structure

Contributing

License

Acknowledgements

Related Documentation

About

Resources

License

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 10

Uh oh!

Contributors

Uh oh!

Languages