Azure Analysis Services MCP Agent

An AI-powered solution that enables natural language interactions with Azure Analysis Services through the Model Context Protocol (MCP), combining a C# Azure Function with a Python MCP server and AI agent.

🏗️ Architecture

This solution uses a hybrid architecture to work around Azure Analysis Services protocol limitations:

┌─────────────────────┐
│   AI Agent/Client   │
│  (uses OpenAI API)  │
└──────────┬──────────┘
           │
           │ MCP Protocol (stdio)
           │
┌──────────▼──────────┐
│    MCP Server       │
│   (Python/stdio)    │
└──────────┬──────────┘
           │
           │ HTTPS
           │
┌──────────▼──────────┐
│  Azure Function     │
│   (.NET 8/C#)       │
│ Managed Identity    │
└──────────┬──────────┘
           │
           │ ADOMD.NET + OAuth2
           │
┌──────────▼──────────┐
│ Azure Analysis      │
│    Services         │
│ (Tabular Model)     │
└─────────────────────┘

Why This Architecture?

Azure Analysis Services requires ADOMD.NET protocol - it does NOT support XMLA over HTTPS. This means:

• ✅ C# with ADOMD.NET works perfectly
• ❌ Python direct connection doesn't work (no native ADOMD library available)
• ❌ Direct MCP → AAS fails
• ✅ Solution: C# Azure Function handles AAS connectivity, Python MCP wraps the HTTP endpoint

For detailed architecture diagrams and decision rationale, see ARCHITECTURE.md.

Components:

1. Azure Function (.NET 8/C#) - Uses ADOMD.NET to connect to AAS with Managed Identity (no secrets!)
2. MCP Server (Python) - Wraps the Azure Function as an MCP tool for AI agents
3. AI Agent (Python) - Uses OpenAI to translate natural language to DAX/MDX queries

✨ Features

• 🔐 Managed Identity - No secrets to manage or rotate (recommended)
• 🚀 Native Protocol - ADOMD.NET for optimal Azure Analysis Services performance
• 🤖 AI-Powered - Natural language queries translated to DAX automatically
• 🔌 MCP Standard - Compatible with any MCP-enabled AI framework
• 📊 Full Query Support - Supports both DAX and MDX queries
• ☁️ Serverless - Azure Functions for scalable, cost-effective deployment
• 🏗️ Infrastructure as Code - Deploy with Bicep or Terraform

📋 Prerequisites

Required Software

• Azure Subscription with Contributor access
• Azure CLI - Download
• .NET 8.0 SDK - Download
• Python 3.10+ - Download
• Azure Functions Core Tools v4 - Download
• Bicep (included with Azure CLI) or Terraform (optional) - Download

Required Azure Resources

• Azure Analysis Services instance with a deployed tabular model (AdventureWorks sample used in this case)
• Azure OpenAI resource for the AI agent
• Administrator access to configure AAS permissions

🚀 Quick Start

1. Clone the Repository

git clone https://github.com/yourusername/azure-aas-mcp-agent.git
cd azure-aas-mcp-agent

2. Deploy the Azure Function

Option 1: Automated Deployment (Recommended)

Using Bicep:

cd infrastructure/bicep
.\deploy.ps1 -Environment dev  # PowerShell
# or
./deploy.sh dev                # Bash

Using Terraform:

cd infrastructure/terraform
terraform init
terraform apply -var-file="dev.tfvars"

Both options automatically:

• Create Azure Function App with Managed Identity
• Configure all required app settings
• Assign permissions to Azure Analysis Services
• Deploy the function code

Option 2: Manual Setup

Follow the comprehensive step-by-step guide:

📖 Complete Setup Guide - Detailed manual deployment instructions

3. Setup the MCP Server

cd mcp-server
pip install -r requirements.txt

# Create .env file
cp .env.example .env

# Edit .env and add your Azure Function URL:
# AZURE_FUNCTION_URL=https://your-function.azurewebsites.net/api/query

4. Run the AI Agent

cd mcp-server
python simple_agent.py

# Ask natural language questions
You: Show me the top 10 products by sales
Agent: [Generates DAX query and returns formatted results]

📁 Project Structure

azure-aas-mcp-agent/
├── README.md                       # This file
├── SETUP_GUIDE.md                  # Step-by-step manual setup instructions
├── ARCHITECTURE.md                 # Detailed architecture diagrams
├── .gitignore                      # Git ignore rules
├── azure-function/                 # C# Azure Function (.NET 8)
│   ├── Program.cs                  # Function host configuration
│   ├── QueryAasFunction.cs         # HTTP trigger (ADOMD.NET client)
│   ├── azure-function.csproj       # Project file with dependencies
│   ├── host.json                   # Function host settings
│   └── local.settings.json.example # Local development settings template
├── infrastructure/                 # Infrastructure as Code
│   ├── README.md                   # Bicep vs Terraform comparison
│   ├── bicep/                      # Azure Bicep templates
│   │   ├── main.bicep              # Main template (Function + Managed Identity)
│   │   ├── parameters.dev.json     # Dev environment parameters
│   │   ├── parameters.prod.json    # Prod environment parameters
│   │   ├── deploy.ps1              # PowerShell deployment script
│   │   ├── deploy.sh               # Bash deployment script
│   │   └── README.md               # Bicep deployment guide
│   └── terraform/                  # Terraform templates
│       ├── main.tf                 # Main configuration
│       ├── variables.tf            # Variable definitions
│       ├── outputs.tf              # Output values
│       ├── dev.tfvars              # Dev environment variables
│       ├── prod.tfvars             # Prod environment variables
│       ├── .gitignore              # Terraform-specific ignores
│       └── README.md               # Terraform guide
└── mcp-server/                     # Python MCP Server & AI Agent
    ├── server.py                   # MCP server (calls Azure Function)
    ├── simple_agent.py             # AI agent example
    ├── requirements.txt            # Python dependencies
    └── .env.example                # Environment variables template

🔧 Configuration

Azure Function App Settings

Automatically configured by Bicep/Terraform, or set manually in Azure Portal:

Note: With Managed Identity, you don't need TENANT_ID, CLIENT_ID, or CLIENT_SECRET!

MCP Server Environment Variables

Create mcp-server/.env from .env.example:

💡 Usage Examples

Using the AI Agent

# Start the agent
cd mcp-server
python simple_agent.py

# Ask natural language questions
You: Show me the top 10 products by sales
Agent: [Generates DAX query and returns formatted results]

You: What are the sales by region?
Agent: [Automatically queries and presents the data]

Direct MCP Tool Usage

from mcp import ClientSession
# ... (see simple_agent.py for full example)

# Call the tool directly
result = await session.call_tool(
    "query_analysis_services",
    arguments={"query": "EVALUATE TOPN(5, 'Product')"}
)

Direct Azure Function Call

curl -X POST https://your-function.azurewebsites.net/api/query \
  -H "Content-Type: application/json" \
  -d '{
    "queryType": "DAX",
    "query": "EVALUATE ROW(\"Test\", 1)"
  }'

🔒 Security Best Practices

• Managed Identity (Recommended): Use system-assigned managed identity for Azure Function → AAS authentication
• Service Principal Permissions: Grant only necessary permissions to Analysis Services
• Function Authentication: Enable Azure Functions authentication for production
• Secrets Management: Use Azure Key Vault for sensitive configuration
• Network Security: Configure firewall rules on Analysis Services
• HTTPS Only: All communication uses HTTPS/TLS encryption

🐛 Troubleshooting

Common Issues

Error: "Unauthorized" when querying Analysis Services

• Verify Managed Identity is added to Analysis Services administrators
• For Service Principal: check format app:{CLIENT_ID}@{TENANT_ID}
• Use Azure Portal → Analysis Services → Server administrators

Error: "Connection timeout"

• Verify Analysis Services firewall allows Azure services
• Check network connectivity from Function App to Analysis Services
• Review Function App networking configuration

Error: "Invalid query"

• Verify DAX syntax is correct
• Check table and column names match your tabular model
• Test query directly in SQL Server Management Studio first

MCP Server connection issues

• Ensure Python dependencies are installed: pip install -r requirements.txt
• Verify AZURE_FUNCTION_URL in .env is correct and accessible
• Check Azure Function logs for errors

Azure Function deployment fails

• Ensure .NET 8.0 SDK is installed
• Verify Azure CLI is authenticated: az login
• Check Azure subscription permissions

For more detailed troubleshooting, see SETUP_GUIDE.md.

📚 Additional Resources

• ARCHITECTURE.md - 11 detailed Mermaid diagrams explaining the architecture
• SETUP_GUIDE.md - Complete step-by-step setup instructions
• infrastructure/README.md - Bicep vs Terraform comparison
• Azure Analysis Services Documentation
• DAX Reference
• Model Context Protocol
• Azure Functions Documentation
• ADOMD.NET Documentation

📞 Support

For questions or issues:

1. Review the SETUP_GUIDE.md and ARCHITECTURE.md
2. Check the troubleshooting section above
3. Search existing issues
4. Create a new issue with detailed information

📄 License

This project is licensed under the MIT License.

Note: This is a proof-of-concept implementation. For production use, consider additional security hardening, error handling, and monitoring capabilities.