> ## Documentation Index
> Fetch the complete documentation index at: https://www.klavis.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Klavis Security

> Comprehensive security layer for MCP integrations protecting against prompt injection, tool poisoning, and other emerging threats.

<img src="https://mintcdn.com/klavisai/ojd5sBgaQMqaHcSS/images/klavis-security/klavis_security.png?fit=max&auto=format&n=ojd5sBgaQMqaHcSS&q=85&s=d218345e82cd15467e5698599280c3dd" alt="Klavis Security" width="1408" height="768" data-path="images/klavis-security/klavis_security.png" />

## Overview

Klavis Guardrails is a comprehensive security layer designed to protect MCP (Model Context Protocol) integrations from emerging threats. It operates as an intelligent proxy between MCP clients and servers, providing real-time threat detection and policy enforcement.

## The Security Challenge

MCP's architecture amplifies security risks by exposing tools, resources, and prompts directly to AI agents. Recent vulnerabilities demonstrate critical flaws:

* **Prompt Injection via Tool Descriptions**: Malicious instructions embedded in MCP tool metadata
* **Cross-Repository Information Leakage**: Agents coerced into accessing private repositories
* **Command Injection and RCE**: Basic security flaws allowing arbitrary code execution
* **Credential Theft**: MCP servers storing OAuth tokens become high-value targets

<img src="https://mintcdn.com/klavisai/ojd5sBgaQMqaHcSS/images/klavis-security/MCP_vulnerabilities.png?fit=max&auto=format&n=ojd5sBgaQMqaHcSS&q=85&s=33bcddaab48e3c7a6414ae4cad19fb6d" alt="MCP Vulnerabilities" width="1536" height="1024" data-path="images/klavis-security/MCP_vulnerabilities.png" />

## Security Architecture

<img src="https://mintcdn.com/klavisai/ojd5sBgaQMqaHcSS/images/klavis-security/klavis_security_architecture.png?fit=max&auto=format&n=ojd5sBgaQMqaHcSS&q=85&s=e407a035cb70f6d69fbcc87ce56e8ea0" alt="Klavis Security Architecture" width="1536" height="1024" data-path="images/klavis-security/klavis_security_architecture.png" />

Klavis Guardrails operates as a security proxy that intercepts, analyzes, and enforces policies on all MCP communication in real-time with four key protection mechanisms:

**Tool Poisoning Detection**: Monitors MCP tool metadata using behavioral analysis to identify when tools deviate from declared functionality.

**Prompt Injection Prevention**: Uses advanced NLP to analyze prompts for malicious instructions, detecting sophisticated attacks before they reach the model.

**Privilege Escalation Monitoring**: Enforces granular access controls ensuring MCP servers operate under least privilege principles.

**Command Injection Mitigation**: Performs deep inspection of tool invocations with strict allowlists and input sanitization.