pc2e-pii-shield

A secure, production-grade Model Context Protocol (MCP) server providing read-only PostgreSQL query execution with automatic, client-side, and edge Personally Identifiable Information (PII) masking. It allows LLM agents (e.g., Cursor, Cline, Claude Code) to execute SQL queries on databases while ensuring strict compliance with GDPR, PDPA, and data privacy principles.

Designed and engineered as a reusable security middleware product, this server intercepts database query results to prevent sensitive data egress.

Technical Architecture

flowchart TD
    Client["AI Agent / Client (Cursor/Cline)"]
    Proxy["Nginx Reverse Proxy"]
    App["pc2e-pii-shield (Express)"]
    DB["Postgres Database (Tailscale-Only)"]

    Client ==>|HTTPS / SSE Request| Proxy
    Proxy ==>|x-api-key Authentication| App
    App ==>|Regex Read-Only Validation| DB
    DB ==>|Raw SQL Results| App
    App ==>|PII Tokenization & Masking| Proxy
    Proxy ==>|Sanitized Event Stream| Client

Core Components

1. Auto-Masking Interceptor (masking.ts): Dynamically scans SQL result sets. It utilizes a hybrid approach: column schema matching (e.g., fields containing name, email, phone) combined with regex-based content scanning to detect and mask sensitive identifiers before data leaves the server.

2. Pseudonymization Cache (cache.ts): An in-memory, TTL-backed cache (default: 30 minutes) that maps raw values to temporary placeholders (e.g., __PERSON_A__, __EMAIL_1__). This allows bi-directional restoration while preventing unbounded memory consumption.

3. AST-Level Mutation Guard (db.ts): A strict regex validator that intercepts raw SQL inputs. It blocks any non-SELECT commands and rejects queries containing forbidden keywords such as DROP, ALTER, DELETE, TRUNCATE, CREATE, or GRANT, ensuring a strict read-only boundary at the application layer.

4. Concurrent Session Manager (index.ts): Unlike basic single-connection templates, this server maintains an active map of SSEServerTransport instances keyed by connection sessionId, allowing multiple remote developers or agents to connect and stream concurrently without state collisions.

5. Telemetry & Metrics Endpoint (/stats): Exposes connection counts, unique client IP tracking, and aggregate query execution statistics to monitor installation and active usage in real-time.

Related MCP server: PostgreSQL MCP Server

Security Model & Threat Mitigation

• Zero-Trust Database Connectivity: Designed to prevent credential exposure. The database runs on an isolated Tailscale-only network interface (e.g., 100.92.174.76), ensuring the database port is never exposed to the public internet.

• Encrypted Transport & API Key Security: The server is fronted by Nginx over HTTPS (port 443) using wildcard SSL certificates, enforcing a secure API key authentication gate (x-api-key) before forwarding requests.

• In-Memory Lifecycle: Pseudonymization mappings are stored in memory with strict TTLs, leaving no persistent disk footprints of the masked PII.

Installation & Deployment

1. Prerequisite Environment Setup

Copy the environment template:

cp .env.example .env

Configure your database credentials and generate a secure API key inside .env.

2. Native Build

Ensure Node.js (v18+) is installed:

npm install
npm run build
npm start

3. Containerized Deployment

Deploy using Docker Compose:

docker compose up -d --build

This maps host port 3088 to the container's internal port 3000, running the SSE server automatically.

4. Direct Execution (NPX)

You can run the server instantly over Stdio transport without downloading the code manually:

npx -y mcp-pii-shield --db-uri "postgresql://username:password@localhost:5432/your_database"

Or run the server over SSE transport:

npx -y mcp-pii-shield --sse --port 3000 --db-uri "postgresql://username:password@localhost:5432/your_database" --api-key "your_secret_key"

Client Integration

A. Local Client Integration (via NPX over Stdio)

Configure your local AI client to launch the server directly using npx.

Claude Desktop (config.json)

Add the following block to your ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pc2e-pii-shield": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-pii-shield",
        "--db-uri",
        "postgresql://username:password@localhost:5432/your_database"
      ]
    }
  }
}

Cursor (Settings → Features → MCP)

1. Click + Add New MCP Server.

2. Set Name to pc2e-pii-shield.

3. Set Type to command.

4. Set Command to:

   npx -y mcp-pii-shield --db-uri "postgresql://username:password@localhost:5432/your_database"

VS Code (Cline / Roo Code)

Add the following to your client settings JSON:

{
  "mcpServers": {
    "pc2e-pii-shield": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-pii-shield",
        "--db-uri",
        "postgresql://username:password@localhost:5432/your_database"
      ]
    }
  }
}

B. Remote Client Integration (via HTTPS over SSE)

If you are connecting to a hosted server (e.g., your public NAS instance), connect via the SSE transport URL.

VS Code (Cline / Roo Code)

{
  "mcpServers": {
    "pc2e-pii-shield": {
      "sseUrl": "https://pii-shield.thegeekybeng.com/sse?api_key=your_api_key_here"
    }
  }
}

Cursor

1. Click + Add New MCP Server.

2. Set Name to pc2e-pii-shield.

3. Set Type to SSE.

4. Set URL to:

   https://pii-shield.thegeekybeng.com/sse?api_key=your_api_key_here

Project Context & Technical Lead

This project was architected, built, and open-sourced by Andrew Yeo.

About the Lead Architect

Andrew is a Senior Systems Architect and AI Engineer based in Singapore, offering:

• 25 years of professional experience in APAC, managing program delivery, client onboarding, and technical vendor management.

• 16+ years of systems architecture and technology leadership, designing and deploying robust enterprise infrastructures and microservice platforms.

• 2+ years of dedicated hands-on AI/ML engineering, specializing in AI safety, LLM metrics, and secure agentic workflows.

Verified Proof-of-Work

• Secure Civic Platforms: Architected and deployed MPS-Connect (a civic constituency casework platform) and Case-Writer-Intelligence (CWI), integrating a 3-stage causality engine with 7 human-in-the-loop approval gates, reducing document triage time by 40%.

• AI Metrology & Testing: Designed the Portable Continuous Context Engine (PC2E), running a systematic, empirical evaluation of 50,000 cases across six LLM providers to benchmark model alignment and compliance.

• Technical Focus: Expert in CI/CD & DevSecOps (GitHub Actions, Docker), containerized deployments, zero-trust network topologies, and local/edge SLM orchestrations.