KindlyGuard

Security-focused MCP (Model Context Protocol) server for AI protection.

Installation

Install KindlyGuard using cargo:

cargo install kindly-guard

This will install the kindly-guard command, which is the actual KindlyGuard server.

What is KindlyGuard?

KindlyGuard is a security-focused MCP server designed to protect AI systems from various threats including:

• Unicode Attacks: Detection and prevention of invisible characters and bidirectional text exploits
• Injection Attempts: Protection against prompt injection and command injection attacks
• Real-time Monitoring: Live threat detection and response
• MCP Protocol Hardening: Security enhancements for the Model Context Protocol

Usage

After installation, you can use KindlyGuard as an MCP server:

# Start the MCP server
kindly-guard --stdio

# Or with a config file
kindly-guard --config /path/to/config.toml

# View all options
kindly-guard --help

Features

• Security Scanner: Advanced threat detection for unicode attacks, injections, and XSS
• Real-time Shield: Live monitoring dashboard showing threat statistics
• Neutralizer: Automatic threat mitigation and sanitization
• Resilience: Built-in circuit breakers and retry logic for fault tolerance
• Storage: SQLite-based persistence for threat history and audit logs

Configuration

KindlyGuard can be configured via a TOML file. Example configuration:

[server]
host = "127.0.0.1"
port = 8080

[scanner]
unicode_enabled = true
injection_enabled = true
xss_enabled = true

[resilience]
[resilience.circuit_breaker]
failure_threshold = 5
recovery_timeout = "30s"

MCP Integration

KindlyGuard implements the Model Context Protocol (MCP) and can be used with any MCP-compatible client. It provides tools for:

• Text scanning and validation
• Security analysis of prompts
• Threat reporting and monitoring

Source Code

The complete KindlyGuard implementation is available at: https://github.com/samduchaine/kindly-guard

License

This project is licensed under the Apache License, Version 2.0.

Author

Created by samduchaine