Enterprise Secure AWS Model Context Protocol (MCP) Server

Production-grade Model Context Protocol (MCP) Server deployed on AWS Lambda & API Gateway with 100% Terraform IaC, zero-trust IAM RBAC, and OWASP LLM Top 10 defense guardrails.

AWS Architecture Overview

Related MCP server: AWS Sage

Features

• Standard MCP Specs: Full support for Anthropic Model Context Protocol (MCP) Tool Discovery (/mcp/tools) and Tool Execution (/mcp/tools/call).

• IAM Role-Based Access Control (RBAC): Restricts tool execution by role (sre_read_only, security_auditor, sre_admin).

• OWASP LLM02 Defense: Edge sanitization blocking subshell metacharacter injections ([;&|'$]).

• OWASP LLM06 Defense: Automatic output buffer redactor scrubbing AWS access keys (AKIA...) and PII.

• Interactive Web UI Playground: Dark-mode test interface running on http://localhost:8080 (make ui).

• Zero Idle Cloud Cost: Provisioned via Terraform with single-command deployment (make deploy) and instant teardown (make destroy).

Web UI Playground Demo

Below is the automated high-definition interactive demonstration showing RBAC access denial, command injection defense, and successful server restart execution under the sre_admin role:

Prerequisites & System Setup

Before setting up the project, ensure your environment has Python 3.11+, Terraform, and AWS CLI installed.

1. System Dependencies (Linux / Ubuntu / Debian)

sudo apt update && sudo apt install python3-venv python3-pip terraform awscli -y

2. Environment Setup

Clone the repository and install project dependencies into an isolated virtual environment (.venv):

git clone https://github.com/patelketul1230/aws-mcp-enterprise-server.git
cd aws-mcp-enterprise-server

# Create .venv and install dependencies
make install

Repository Structure

aws-mcp-enterprise-server/
├── docs/
│   ├── assets/                    # Centralized PNG images & HD animations
│   │   ├── mcp_aws_figma_architecture_diagram.png
│   │   ├── mcp_ui_testing_demo.webp
│   │   ├── mcp_ui_testing_demo.gif
│   │   ├── mcp_ui_playground_success_screenshot.png
│   │   ├── mcp_ui_playground_admin_success_screenshot.png
│   │   └── mcp_ui_playground_blocked_screenshot.png
│   ├── blogs/                     # 3 Medium Sub-Blogs ready for publication
│   │   ├── blog_1_1_mcp_decoded.md
│   │   ├── blog_1_2_mcp_rbac_python.md
│   │   └── blog_1_3_mcp_tool_poisoning.md
│   ├── architecture.md            # Figma diagrams & sequence flows
│   ├── 5w_and_how.md              # 5 Ws + 1 H problem statement & threat model
│   └── demo_execution.log         # Verified test execution log
├── src/
│   ├── clients/                   # AWS Bedrock Converse API integration client
│   ├── middleware/                # MCPSecurityValidator (RBAC & DLP)
│   ├── tools/                     # CloudWatch & Terraform tool modules
│   ├── ui/                        # Interactive Web UI Playground (index.html)
│   └── server.py                  # AWS Lambda & FastAPI server handler
├── terraform/                     # 100% Terraform IaC (API Gateway, IAM, Lambda)
├── tests/                         # Pytest suite & live verification scripts
├── Makefile                       # Lifecycle hooks (make deploy / make destroy / make ui)
└── README.md                      # Project documentation

Quick Start Guide

1. Install Dependencies

make install

2. Run Tests

make test

3. Deploy Infrastructure to AWS

make deploy

Outputs:

• API Endpoint: Dynamic output generated by Terraform

• CloudWatch Log Group: /aws/mcp/aws-mcp-enterprise-server-dev

4. Launch Interactive Web UI Playground

make ui

Open http://localhost:8080 in your web browser to test tool calls, RBAC roles, and injection defenses visually!

5. Test via Terminal (curl)

# Health Check
curl -s <YOUR_API_ENDPOINT>/health

# Tool Discovery Catalog
curl -s <YOUR_API_ENDPOINT>/mcp/tools

# Valid Tool Call (CloudWatch Query)
curl -s -X POST <YOUR_API_ENDPOINT>/mcp/tools/call \
  -H "Content-Type: application/json" \
  -H "X-User-Role: sre_read_only" \
  -d '{"tool_name": "query_cloudwatch_logs", "arguments": {"log_group": "/aws/lambda/payment-dev", "filter_pattern": "ERROR"}}'

# Command Injection Attack (Returns 403 Forbidden)
curl -s -X POST <YOUR_API_ENDPOINT>/mcp/tools/call \
  -H "Content-Type: application/json" \
  -H "X-User-Role: sre_read_only" \
  -d '{"tool_name": "query_cloudwatch_logs", "arguments": {"log_group": "/aws/lambda/payment-dev; rm -rf /", "filter_pattern": "ERROR"}}'

6. Tear Down AWS Resources

make destroy

Technical Documentation Links

• System Architecture & Sequence Flows

• 5 Ws + 1 H Architectural Breakdown

• Medium Sub-Blog 1.1: MCP Decoded

• Medium Sub-Blog 1.2: IAM RBAC in Python

• Medium Sub-Blog 1.3: Tool Poisoning Defenses

License & Enterprise Inquiries

This project is Dual-Licensed:

• Open Source / Non-Commercial Use: Licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). Free for open-source projects, personal evaluation, and educational use.

• Enterprise / Commercial Production Use: Requires a paid Enterprise Commercial License. If your company intends to deploy or integrate this MCP server inside proprietary commercial applications, SaaS platforms, or private cloud infrastructure without open-sourcing the surrounding codebase, please contact the author for licensing terms.

Licensing Requests & Inquiries:

• Author & Copyright Holder: Ketul Patel

• Licensing Request Email: kpsub786@gmail.com