The original title is "Turning a System of Record into an AI Agent: Building MCP Tools on Azure"

Short summary

A detailed walkthrough of exposing a read-only enterprise system of record to an AI agent using Azure Logic Apps, API Management, and Agent Foundry via Model Context Protocol (MCP). The architecture uses a single backend endpoint fanned out into many named tool operations, with constraints around read-only access, prompt-injection safety, and composability. Implementation notes cover nested switch patterns, secret management, and preferring server-side aggregates over client-side counting.

• •Five-layer architecture: Agent → MCP server → API gateway → tool executor → source system, all read-only
• •One backend endpoint fans out into many typed MCP tools via API Management
• •Practical guidance on nested switches, Key Vault secrets, escaping @ symbols, and seeding default error responses

Generated with AI, which can make mistakes.