Case Study: Customer Data Platform

Enterprise CRM with MCP

A B2B SaaS company built an internal AI assistant that connects to their customer data platform via MCP. The assistant handles 500+ customer queries per day from the sales and support teams.

Architecture

┌─────────────────────────────────────────────┐
│        INTERNAL CHAT APP (Custom MCP Host)   │
├─────────────────────────────────────────────┤
│  MCP Gateway (central proxy)                 │
│  ├── CRM Server (Salesforce data)            │
│  ├── Analytics Server (Mixpanel events)      │
│  ├── Billing Server (Stripe data)            │
│  ├── Support Server (Zendesk tickets)        │
│  └── Knowledge Base Server (Confluence)      │
├─────────────────────────────────────────────┤
│  Security Layer:                             │
│  • OAuth 2.1 per server                      │
│  • Role-based tool access                    │
│  • Full audit logging                        │
│  • PII redaction on responses                │
└─────────────────────────────────────────────┘

Role-Based Access Control

PII Redaction Pattern

// MCP Gateway middleware: redact PII before returning to LLM
function redactPII(response: ToolResult, userRole: string): ToolResult {
  if (userRole === "intern" || userRole === "external") {
    const text = response.content[0].text;
    return {
      content: [{
        type: "text",
        text: text
          .replace(/\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/gi, "[EMAIL REDACTED]")
          .replace(/\b\d{3}[-.]?\d{3}[-.]?\d{4}\b/g, "[PHONE REDACTED]")
          .replace(/\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b/g, "[CARD REDACTED]")
      }]
    };
  }
  return response;
}

Key Results