GuardrailAI

Autonomous Regulatory Guardrail Agent using MCP and Google Gemini

GuardrailAI is an AI-powered compliance agent that evaluates file write requests before execution. By combining deterministic policy validation with Google Gemini reasoning, it detects sensitive information, explains compliance decisions, and prevents insecure data storage through an auditable governance workflow.

Dashboard Preview

AI Governance Dashboard

Live Audit Timeline

Problem

AI-powered applications frequently generate and write sensitive information such as passwords, API keys, personally identifiable information (PII), and confidential business data. Traditional rule-based validation lacks contextual reasoning and explainability, making governance and compliance difficult.

GuardrailAI ensures every file write request is evaluated before execution, reducing security risks while providing transparent, explainable decisions.

Solution

GuardrailAI processes every request through an AI-driven compliance pipeline:

• Policy-based security validation

• Sensitive data detection

• Google Gemini compliance reasoning

• Risk score calculation

• Automated approval or rejection

• Immutable audit logging

• Live governance dashboard

System Architecture

Technology Stack

Key Features

• MCP-based agent workflow

• Google Gemini compliance reasoning

• Multi-agent architecture

• Policy-driven validation

• API key and secret detection

• PII detection

• Password detection

• Risk scoring

• Explainable AI decisions

• Interactive governance dashboard

• Audit log generation

• Compliance report export

Note : The MCP server communicates via the Model Context Protocol (STDIO transport) and is intended to be run locally with an MCP-compatible client. The deployed web application hosts the dashboard interface and visualization layer.

Project Structure

guardrail_ai/
│
├── agents/
├── api/
├── config/
├── dashboard/
├── logs/
├── output/
├── tools/
├── utils/
├── server.js
├── dashboardServer.js
└── package.json

Setup

Clone the repository

git clone https://github.com/<username>/GuardrailAI.git

Navigate into the project

cd GuardrailAI

Install dependencies

npm install

Create a .env file

GEMINI_API_KEY=YOUR_API_KEY

Run the application

npm start

Open

http://localhost:3000

Concepts Demonstrated

• Model Context Protocol (MCP)

• Multi-Agent System

• Google Gemini Integration

• Security-Focused AI Agent

• Explainable AI

• Deployable Web Application

Future Enhancements

• Policy Management Interface

• Role-Based Access Control

• PDF Compliance Reports

• Historical Compliance Analytics

• Multi-user Support

License

MIT License