Shrike

Acerca de

Shrike is an MCP (Model Context Protocol) server designed to safeguard AI agents against a wide range of security threats. It acts as a critical layer, scanning user prompts for malicious content such as prompt injection, jailbreaks, SQL injection attempts, and PII exposure, while also validating LLM-generated responses for system prompt leaks, unexpected PII, and toxic language. With features like PII redaction and token-based rehydration, it ensures data privacy and compliance. Shrike implements a robust 'fail-closed' security model, blocking requests on network timeouts or backend errors to prevent bypass attacks, making it an essential tool for secure AI agent deployment.

Características Principales

• Scans LLM responses for system prompt leaks, PII, and toxic language.
• Validates SQL queries and file write operations to prevent malicious actions.
• Comprehensive threat detection for AI agents (prompt injection, jailbreak, SQL injection, data exfiltration).
• 0 GitHub stars
• Implements a fail-closed security model to block requests on errors or timeouts, preventing bypasses.
• PII redaction from prompts and rehydration into responses for enhanced privacy.

Casos de Uso

• Securing AI applications and chatbots against user-driven prompt attacks and data exfiltration.
• Ensuring privacy and compliance by redacting and rehydrating PII in AI interactions.
• Protecting backend systems by validating AI-generated SQL queries and file operations.