How Block Revolutionized E-Commerce Support with Custom MCP Servers

Executive Summary

The Challenge: Scaling Support for Millions of Merchants

In late 2024, Block faced a critical challenge. With over 4 million active merchants processing billions in transactions, their customer support infrastructure was overwhelmed. The company was handling:

• 50,000+ daily support queries across email, chat, and phone
• Average response time of 48 hours for complex issues
• 20% annual growth in support volume
• Disconnected data across 12 different systems
• High agent turnover due to repetitive query handling

"We knew AI could help, but traditional chatbots weren't cutting it. They couldn't access our systems, couldn't understand context, and definitely couldn't resolve complex merchant issues." - Sarah Chen, VP of Customer Experience at Block

The Solution: Custom MCP Server Architecture

Phase 1: Foundation (Weeks 1-4)

Block's engineering team began by implementing the Model Context Protocol to create a unified communication layer between their AI assistants and backend systems.

Technical Architecture

// Example: Order Management MCP Server
class OrderManagementMCP {
  constructor(private apiClient: BlockAPIClient) {
    this.setupTools();
  }

  @mcpTool("get_order_details")
  async getOrderDetails(orderId: string) {
    // Fetch from multiple systems
    const order = await this.apiClient.orders.get(orderId);
    const payment = await this.apiClient.payments.getByOrder(orderId);
    const shipping = await this.apiClient.shipping.track(order.trackingId);
    
    return {
      order,
      payment,
      shipping,
      context: this.generateContext(order)
    };
  }

  @mcpTool("process_refund")
  async processRefund(orderId: string, reason: string) {
    // Multi-step refund process with validation
    const validation = await this.validateRefund(orderId);
    if (validation.approved) {
      return await this.apiClient.refunds.create({
        orderId,
        reason,
        autoApproved: true
      });
    }
  }
}

Phase 2: Integration (Weeks 5-8)

The team integrated MCP servers with their existing Claude-powered support system, enabling context-aware responses:

• Order History Access: AI could retrieve complete transaction history
• Real-time Inventory: Connected to warehouse systems for stock information
• Payment Processing: Enabled secure refund and dispute handling
• Merchant Profiles: Access to account settings and preferences

Implementation Deep Dive

Custom Tools Developed

Security and Compliance

Block implemented enterprise-grade security measures:

• End-to-end encryption for all MCP communications
• PCI DSS compliance for payment data handling
• GDPR-compliant data access controls
• Comprehensive audit logging for all AI actions
• Rate limiting to prevent system abuse

Results: Transformative Impact

"The MCP implementation didn't just improve our metrics – it fundamentally transformed how we think about customer support. Our agents are now advisors, not ticket processors." - Michael Rodriguez, Director of Engineering at Block

Key Learnings and Best Practices

What Worked Well

1. Incremental Rollout: Starting with low-risk queries before expanding to complex issues
2. Human-in-the-Loop: Maintaining oversight for sensitive operations like refunds
3. Continuous Learning: Using failed queries to improve MCP tool design
4. Cross-Team Collaboration: Engineering and support teams working together from day one

Challenges Overcome

• Legacy System Integration: Some APIs required wrapper services for MCP compatibility
• Data Consistency: Implemented caching strategies to handle system latency
• Change Management: Required extensive training for support agents
• Scaling Issues: Initial architecture needed optimization for peak loads

Technical Insights

Future Roadmap

Block's success with MCP has led to expanded plans for 2025: