ShieldDB 🛡️

The Secured Google Gemini Database Workstation & MCP Gateway

🌐 Live Dashboard · ▶ Watch Demo

---

ShieldDB is an enterprise-grade Model Context Protocol (MCP) gateway, security sidecar, and interactive workstation that enables Google Gemini to securely administer databases. Powered by gemini-2.5-flash, the workstation translates plain English instructions into complex database operations. To ensure enterprise compliance and prevent malicious exploits, all Gemini database interactions are intercepted and audited by DuoGuard-0.5B — a state-of-the-art multilingual safety classifier that blocks inbound prompt injections and automatically redacts outbound customer PII.

---

🌟 Key Features

1. Autonomous Gemini Database Workstation 🤖

   • Employs Google Gemini (gemini-2.5-flash) as the primary cognitive engine.
   • Translates complex natural language prompts (e.g., "Give me a summary of active users in NY and update their status to active") into safe, structured database commands via Model Context Protocol (MCP) tool bindings.

2. Inbound Guardrails (Safety Shielding) 🛡️

   • Intercepts Gemini's incoming instructions to protect database collections against prompt injections, bypass attempts, and destructive command injections.
   • Detects risks across 12 distinct safety categories (Jailbreak prompts, Non-violent crimes, Violent crimes, Suicide, etc.) in milliseconds.

3. Outbound Guardrails (PII & Privacy Redaction) 🔒

   • Automatically sanitizes retrieved query results before sending them to Gemini or rendering them in the chat workspace.
   • Intelligently redacts Emails, Credit Cards, Social Security Numbers, Phone Numbers, Passwords, and API Keys.
   • Enables full compliance with strict privacy standards like GDPR, HIPAA, and PCI-DSS.

4. Smart Hybrid Database Engine ⚙️

   • Automatically detects live MongoDB Atlas credentials in the environment.
   • Falls back to an In-Memory Sandbox (MongoMock) if offline, pre-seeding it with rich customer records so judges can run and evaluate the system instantly.

5. Premium Security & Chat Dashboard 💎

   • Includes the Gemini Secure Brain workspace — a clean, real-time interactive chat dashboard to instruct Gemini in plain English.
   • Features a Live Security Console & Playground to test benign vs. malicious prompts and visualize DuoGuard risk profiles and query redactors in real-time.

---

🛠️ Architecture

flowchart TD
    A[User Prompt] --> B[Google Gemini\ngemini-2.5-flash]
    B -->|Secure MCP Tool Call| C[ShieldDB MCP Gateway]
    C --> D[Inbound Screening\nDuoGuard-0.5B]
    C --> E[Database Adapter\nMongoDB / MongoMock]
    D -->|Block injections / jailbreaks| C
    E --> F[Outbound Redaction\nScrub Emails / CC / PII]
    F --> G[Sanitized Query Results]
    G --> B
    B --> H[Safe Natural Response]

Loading

---

🚀 Installation & Local Startup

Ensure you have Python 3.10+ and the uv package manager installed.

0. Environment Setup

To enable the interactive Gemini chat agent workspace, set your Google Gemini API key as an environment variable (or create a .env file in the root directory):

export GEMINI_API_KEY="your_gemini_api_key_here"

1. Pre-download Safety Models

To ensure instant startup, pre-cache the DuoGuard weights and tokenizers locally:

uv run python -m duogaurd_mcp.main download

2. Start the Security Dashboard & API Backend

Launch the FastAPI REST API serving telemetry and playgrounds:

uv run python -m duogaurd_mcp.main dashboard

The API will be available at http://localhost:8000.

3. Run the MCP Server (stdio)

To connect ShieldDB to Claude Desktop or Cursor IDE, run:

uv run python -m duogaurd_mcp.main run

---

📂 Project Structure

duogaurd_mcp/
├── pyproject.toml              # UV dependency metadata
├── README.md                   # System documentation
├── duogaurd_mcp/
│   ├── __init__.py
│   ├── moderator.py            # Safety classification & PII regex redactor
│   ├── database.py             # MongoDB connection & sandbox seed generator
│   ├── server.py               # Secure FastMCP tools
│   ├── api.py                  # FastAPI REST endpoints
│   ├── main.py                 # CLI controller (run, dashboard, download)
│   └── test.py                 # Diagnostic test suite
└── dashboard/                  # React + TypeScript + Vite frontend

---

🛡️ License

Open-Source under the MIT License. Built for the Google Cloud Rapid Agent Hackathon.