Azure Functions

Development guidance for Azure Functions: triggers, bindings, Durable Functions, local dev, deployment, and MCP integration.

When to Use

• Implementing event-driven workloads (HTTP, queue, timer, blob, Service Bus, Event Grid triggers)
• Building Durable Functions orchestrations (fan-out/fan-in, eternal orchestrations, human approval)
• Setting up local development with func CLI and Azurite
• Choosing a hosting plan (Consumption vs Premium vs Dedicated)
• Deploying Functions via az, azd, or CI/CD pipelines
• Exposing Azure Functions as MCP tools for AI agents

Quick Start

/azure-functions triggers    # Trigger types, binding input/output patterns
/azure-functions durable     # Durable Functions: orchestrators, activities, patterns
/azure-functions local       # Local dev: func CLI, Azurite, local.settings.json
/azure-functions deploy      # Hosting plan selection, deployment, slots
/azure-functions mcp         # MCP extension: expose functions as AI tools

Trigger and Binding Overview

Output Bindings (declarative, no SDK required)

Input trigger → Function logic → Output bindings

Output bindings: Service Bus message, Blob, Cosmos DB document, SignalR, Event Hubs, Queue Storage, Durable activity scheduling.

Local Development

Prerequisites:

• Azure Functions Core Tools: npm install -g azure-functions-core-tools@4 --unsafe-perm true
• Azurite (local Storage emulator): npm install -g azurite
• Language runtime for your target (Node.js, Python, Go, .NET)

# Initialize a new Functions project
func init MyFunctionApp --worker-runtime node     # or python, dotnet, custom
cd MyFunctionApp

# Create a function
func new --name HttpTrigger --template "HTTP trigger"
func new --name TimerTrigger --template "Timer trigger"
func new --name ServiceBusTrigger --template "Azure Service Bus trigger"

# Start Azurite (Storage emulator) in a separate terminal
azurite --location ~/.azurite --debug ~/.azurite/debug.log

# Start the Functions host
func start

# Run a specific function
func run HttpTrigger

local.settings.json

{
  "IsEncrypted": false,
  "Values": {
    "AzureWebJobsStorage": "UseDevelopmentStorage=true",
    "FUNCTIONS_WORKER_RUNTIME": "node",
    "MyServiceBusConnection": "Endpoint=sb://...",
    "MyCosmosConnection": "AccountEndpoint=https://..."
  }
}

Never commit local.settings.json — add to .gitignore. Use Key Vault references or environment-specific config for production values.

See references/local-development.md for debugging, remote attach, and Azurite configuration.

Hosting Plan Selection

2025 Recommendation: Flex Consumption is the preferred plan for new production deployments — it combines Consumption pricing with configurable minimum instances (eliminating cold starts when needed) and 60-minute max duration.

Deployment

# Deploy via Azure CLI
az functionapp deployment source config-zip \
  --resource-group <rg> \
  --name <function-app> \
  --src build.zip

# Deploy via func CLI
func azure functionapp publish <function-app>

# Deploy via azd
azd deploy

# List deployed functions
az functionapp function list --name <function-app> --resource-group <rg> --output table

Deployment Slots

Deployment slots enable zero-downtime releases with instant swap rollback.

# Create a staging slot
az functionapp deployment slot create \
  --name <function-app> \
  --resource-group <rg> \
  --slot staging

# Deploy to staging
az functionapp deployment source config-zip \
  --resource-group <rg> \
  --name <function-app> \
  --slot staging \
  --src build.zip

# Swap staging → production (atomic)
az functionapp deployment slot swap \
  --resource-group <rg> \
  --name <function-app> \
  --slot staging \
  --target-slot production

# Rollback: swap back
az functionapp deployment slot swap \
  --resource-group <rg> \
  --name <function-app> \
  --slot production \
  --target-slot staging

See references/deployment-patterns.md for CI/CD pipeline integration, slot-specific settings, and traffic splitting.

Azure Functions MCP Extension

The Azure Functions MCP extension (GA) allows HTTP-triggered functions to be exposed as MCP tools — enabling AI agents to invoke functions directly.

// host.json — enable MCP extension
{
  "version": "2.0",
  "extensionBundle": {
    "id": "Microsoft.Azure.Functions.ExtensionBundle",
    "version": "[4.0.0, 5.0.0)"
  },
  "extensions": {
    "mcp": {
      "instructions": "You are a helpful assistant. Use these tools to complete tasks."
    }
  }
}

// TypeScript — HTTP function exposed as MCP tool
import { app, HttpRequest, HttpResponseInit, InvocationContext } from "@azure/functions";

app.http("listOrders", {
  methods: ["GET"],
  authLevel: "function",
  handler: async (request: HttpRequest, context: InvocationContext): Promise<HttpResponseInit> => {
    const customerId = request.query.get("customerId");
    // ... business logic
    return { jsonBody: orders };
  },
  // MCP tool metadata
  extraInputs: [],
  extraOutputs: [],
});

Auth: The MCP extension uses On-Behalf-Of (OBO) auth — the calling user's token is forwarded. Requires Entra ID app registration with appropriate API permissions.

Transport: Streamable HTTP (SSE). Compatible with Claude, Claude Desktop, and Cursor.

See references/triggers-and-bindings.md for binding code examples per language.

Cold Start Mitigation

• Flex Consumption: Set minimumInstanceCount > 0 to keep warm instances.
• Premium: Always-ready instances; no cold start.
• Consumption with WEBSITE_RUN_FROM_PACKAGE=1: Reduces startup time by running from a mounted zip.
• Keep function packages small: Remove unused dependencies; tree-shake Node.js bundles.
• Avoid heavy initialization at module level: Move SDK client creation to lazy initialization inside the handler or use module-level singletons carefully.

Key References