Secure GenAI Gateway & MCP Server

🤖 Project Overview

This project implements a Secure GenAI Gateway that acts as a centralized control plane for accessing Large Language Models (LLMs) like Amazon Bedrock. It enforces security policies, sanitizes inputs to prevent PII leakage, and implements Model Context Protocol (MCP) patterns to route requests effectively.

Key Features:

Centralized Gateway: Single API entry point for all internal AI consumption.
Input Sanitization: Python-based middleware to strip PII and block prompt injection attacks.
Model Context Protocol (MCP): Structured routing to provide LLMs with necessary context and tools.
Private Access: Configured with Resource Policies to restrict access to Corporate VPNs.
Serverless Architecture: Built on AWS Lambda and API Gateway for infinite scaling and zero idle cost.

🏗️ Architecture

API Gateway: Handles authentication and rate limiting.
Lambda Middleware: Executes sanitization logic and MCP routing.
Amazon Bedrock: Provides access to foundation models (Claude 3, Titan, etc.).

🛠️ Tech Stack

Compute: AWS Lambda (Python 3.9)
API: Amazon API Gateway (REST)
AI Model: Amazon Bedrock (Claude 3 Sonnet)
IaC: AWS SAM (Serverless Application Model)

📂 Repository Structure

├── src/
│   ├── lambda_function.py  # Core Logic (Sanitization + Bedrock Call)
│   └── requirements.txt    # Dependencies
├── template.yaml           # AWS SAM Infrastructure Definition
└── GenAI_Project_Guide.md  # Detailed Documentation

🔒 Security Features

PII Redaction: Automatically detects and blocks sensitive data patterns (SSN, Email).
Prompt Firewall: Prevents "Jailbreak" attempts via system prompt enforcement.
Audit Logging: Full request/response tracing in CloudWatch Logs.

🚀 Deployment

To deploy this stack using AWS SAM:

sam build
sam deploy --guided

📄 License

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

MCP Servers