Model Context Protocol (MCP) has become the standard way to expose tools to AI agents.

With Azure Logic Apps, you can create and run multiple MCP servers and let an agent consume them together — cleanly and modularly.

In this post, we'll build:

• A Basic Arithmetic MCP server
• An Extended Arithmetic MCP server
• And connect an agent to both

The Scenario

We'll create two MCP servers using Logic App workflows and expose them to an agent.
Both servers share Anonymous authentication.

Step 1 — Create and Group MCP Workflows in Logic Apps

Each operation is implemented as a Logic App workflow and exposed as an MCP tool.

Each workflow:

• Accepts inputs (typically two numbers and an operation)
• Executes the required logic
• Returns a structured response

You then group these workflows into MCP servers:

Basic Arithmetic Server

• Add
• Subtract
• Multiply
• Divide

Extended Arithmetic Server

• Power
• Square Root
• Modulo

How This Is Stored (mcpservers.json)

Once workflows are grouped into MCP servers, the configuration is automatically persisted in the mcpservers.json file.

This file contains:

• MCP server definitions
• Workflow (tool) mappings

💡 Key idea: What you define as MCP servers (grouped workflows) is what gets written to mcpservers.json — automatically.

MCP Server Endpoints

Once registered, each MCP server is reachable at a predictable URL following this pattern:

https://<your-logic-app>.azurewebsites.net/api/mcpservers/<ServerName>/mcp

For example:

• Basic server → /api/mcpservers/BasicArithmetic/mcp
• Extended server → /api/mcpservers/ExtendedArithmetic/mcp

💡 The server name in the URL matches exactly what you defined when grouping your workflows — no extra configuration needed.

Step 2 — Connect the Agent to Both Servers

The agent connects to both MCP servers using separate MCP client connections.

This allows the agent to:

• Use basic operations from one server
• Use advanced operations from another

👉 Clean separation — no need for a single large service.

Step 3 — Execution Result

When the agent runs, it invokes operations across both MCP servers.

The result:

• Multiple MCP servers used seamlessly
• Operations resolved correctly
• Agent behaves as if using a unified toolset

Why This Matters

Running multiple MCP servers from a single Logic Apps instance gives you separation of concerns, modular extensibility, and independent scalability — without changing your agent design.

Key Takeaway

• Workflows group into MCP servers, each exposed via a predictable endpoint
• Configuration is automatically managed in mcpservers.json
• Agents can connect to multiple MCP servers simultaneously — no extra wiring required

This pattern enables a modular, composable tool ecosystem for AI agents using Azure Logic Apps.