Microsoft Fabric Analytics MCP Server

A comprehensive Model Context Protocol (MCP) server that provides analytics capabilities and tools for interacting with Microsoft Fabric data platform. This server enables AI assistants like Claude to seamlessly access, analyze, and monitor Microsoft Fabric resources through standardized MCP protocols, bringing the power of Microsoft Fabric directly to your AI conversations.

📋 Table of Contents

• 🌟 Key Features

• 🚀 Quick Start

• 🛠️ Tools & Capabilities

• 🧪 Development & Testing

• 💬 Example Queries

• 🔐 Authentication

• 🏗️ Architecture

• ⚙️ Configuration

• 🤝 Contributing

• 🔒 Security "Cancel the problematic Spark application"

• 📞 Support

🌟 Key Features

• 🏗️ Complete Workspace Management - Create, delete, and manage Fabric workspaces with capacity assignment

• 🔄 Enhanced CRUD Operations - Create, read, update, and delete all Fabric items (notebooks, lakehouses, datasets, reports)

• 📓 Advanced Notebook Management - Create, execute, and manage Fabric notebooks with 5 predefined templates

• ⚡ Livy API Integration - Full Spark session and batch job management with real-time monitoring

• 📊 Comprehensive Spark Monitoring - Real-time monitoring across workspaces, items, and applications

• 🤖 Multi-AI Assistant Support - Works with Claude Desktop, GitHub Copilot, and other MCP-compatible AI tools

• 🔐 Enhanced Azure CLI Authentication - Zero-config setup with automatic token management

• �️ Enterprise Authentication - Multiple auth methods (Bearer, Service Principal, Device Code, Interactive, Azure CLI)

• 📈 Analytics & Insights - Generate comprehensive monitoring dashboards with real-time metrics

• 🧪 End-to-End Testing - Complete test suite with real workspace creation and job execution

• 🔄 Advanced Token Management - Automatic token validation, refresh, and expiration handling "List all Fabric capacities I can use" "Assign workspace 1234abcd-abcd-1234-abcd-123456789000 to capacity f9998888-7777-6666-5555-444433332222" "Show all workspaces in capacity f9998888-7777-6666-5555-444433332222" "Unassign workspace 1234abcd-abcd-1234-abcd-123456789000 from its capacity"

"Create a new workspace called 'Analytics-Q1-2025' and assign it to our premium capacity"

"List all workspaces in our tenant and show their capacity assignments"

"Add user john.doe@company.com as Admin to the Analytics workspace"

"Create a development environment in the Analytics workspace with Python and R libraries"

"Connect the Analytics workspace to our GitHub repository for version control"

🔀 Synapse to Fabric Migration Tools

🆕 Automated Spark Workload Migration

The MCP server now includes 4 specialized migration tools that automate the migration of Spark notebooks and pipelines from Azure Synapse Analytics to Microsoft Fabric:

🔍 Migration Discovery Tools

• fabric_list_synapse_workspaces - List all Synapse workspaces in your Azure subscription

• fabric_discover_synapse_workspace - Inventory notebooks, pipelines, linked services, and Spark jobs from Synapse

🔄 Transformation & Migration Tools

• fabric_transform_notebooks - Transform Synapse notebooks to Fabric format (mssparkutils → notebookutils)

• fabric_migrate_synapse_to_fabric - Complete end-to-end migration with discovery, transformation, and provisioning

✨ Key Migration Features

• Automatic Code Transformation - Converts Synapse-specific code to Fabric equivalents:

  • mssparkutils → notebookutils

  • Synapse magic commands → Fabric magic commands

  • ABFSS path rewriting to OneLake

  • Spark pool configuration cleanup

• Comprehensive Asset Discovery - Inventories all migrat assets:

  • Jupyter notebooks (ipynb format)

  • Data pipelines and workflows

  • Linked services and connections

  • Spark job definitions

• Safe Testing with Dry Run - Preview all changes before applying:

  • Test transformations without provisioning

  • Validate transformed code

  • Review change reports

• End-to-End Automation - Complete migration pipeline:

  • Discovery → Transformation → Provisioning → Validation

  • Automatic lakehouse creation

  • OneLake shortcut provisioning

  • Comprehensive migration reports

🎯 Migration Scenarios

📋 Explore Before Migrating:

🔄 Preview Transformations:

🚀 Complete Migration:

📊 Detailed Migration Guide: See MIGRATION.md for comprehensive migration documentation including:

• Step-by-step workflows

• Transformation rule details

• Best practices and troubleshooting

• Complete examples

🎯 End-to-End Testing with Real Workspaces

The MCP server now includes comprehensive end-to-end testing that creates real workspaces, assigns them to capacities, and executes actual jobs to validate the complete workflow:

What it tests:

• ✅ Workspace Creation - Creates real Fabric workspaces

• ✅ Capacity Assignment - Attaches workspaces to your Fabric capacity

• ✅ Item Creation - Creates notebooks, lakehouses, and other items

• ✅ Job Execution - Runs actual Spark jobs and monitors completion

• ✅ Resource Cleanup - Automatically removes all test resources

🚀 Deployment Options

🤖 Claude Desktop Integration

Recommended for AI Assistant Usage:

💡 Get Bearer Token: Visit Power BI Embed Setup to generate tokens

⚠️ Important: Tokens expire after ~1 hour and need to be refreshed

🔧 Claude Desktop Authentication Fix

If you experience 60-second timeouts during startup, this is due to interactive authentication flows blocking Claude Desktop's sandboxed environment. Solution:

1. Use Bearer Token Method (Recommended):

   • Set FABRIC_AUTH_METHOD: "bearer_token" in your config

   • Provide FABRIC_TOKEN with a valid bearer token

   • This bypasses interactive authentication entirely

2. Alternative - Per-Tool Authentication:

   • Provide token directly in tool calls: bearerToken: "your_token_here"

   • Or use simulation mode: bearerToken: "simulation"

3. Troubleshooting:

   • Server now has 10-second timeout protection to prevent hanging

   • Falls back to simulation mode if authentication fails

   • Enhanced error messages provide clear guidance

🎯 Quick Fix: The server automatically prioritizes FABRIC_TOKEN environment variable over interactive authentication flows, preventing Claude Desktop timeouts.

📱 Local Development

🐳 Docker Deployment

☸️ Azure Kubernetes Service (AKS)

🌐 Azure MCP Server (Preview)

📚 Detailed Guides:

• 🐳 Docker & Compose Setup

• ☸️ AKS Deployment Guide

• 🌐 Azure MCP Server Guide

• 🔧 Configuration Examples

• ✅ Deployment Validation

🛠️ Tools & Capabilities

🔍 CRUD Operations for Fabric Items

• Tool: list-fabric-items

• Description: List items in a Microsoft Fabric workspace (Lakehouses, Notebooks, etc.)

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • itemType: Filter by item type (optional)

• Tool: create-fabric-item

• Description: Create new items in Microsoft Fabric workspace

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • itemType: Type of item (Lakehouse, Notebook, Dataset, Report, Dashboard)

  • displayName: Display name for the new item

  • description: Optional description

• Tool: get-fabric-item

• Description: Get detailed information about a specific Microsoft Fabric item

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • itemId: ID of the item to retrieve

• Tool: update-fabric-item

• Description: Update existing items in Microsoft Fabric workspace

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • itemId: ID of the item to update

  • displayName: New display name (optional)

  • description: New description (optional)

• Tool: delete-fabric-item

• Description: Delete items from Microsoft Fabric workspace

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • itemId: ID of the item to delete

🔍 Query Fabric Dataset (Enhanced)

• Tool: query-fabric-dataset

• Description: Execute SQL or KQL queries against Microsoft Fabric datasets

• Parameters:

  • bearerToken: Microsoft Fabric bearer token (optional - uses simulation if not provided)

  • workspaceId: Microsoft Fabric workspace ID

  • datasetName: Name of the dataset to query

  • query: SQL or KQL query to execute

🚀 Execute Fabric Notebook

• Tool: execute-fabric-notebook

• Description: Execute a notebook in Microsoft Fabric workspace

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • notebookId: ID of the notebook to execute

  • parameters: Optional parameters to pass to the notebook

📊 Get Analytics Metrics

• Tool: get-fabric-metrics

• Description: Retrieve performance and usage metrics for Microsoft Fabric items

• Parameters:

  • workspaceId: Microsoft Fabric workspace ID

  • itemId: Item ID (dataset, report, etc.)

  • timeRange: Time range for metrics (1h, 24h, 7d, 30d)

  • metrics: List of metrics to analyze

🔧 Analyze Data Model

• Tool: analyze-fabric-model

• Description: Analyze a Microsoft Fabric data model and get optimization recommendations

• Parameters:

  • workspaceId: Microsoft Fabric workspace ID

  • itemId: Item ID to analyze

📋 Generate Analytics Report

• Tool: generate-fabric-report

• Description: Generate comprehensive analytics reports for Microsoft Fabric workspaces

• Parameters:

  • workspaceId: Microsoft Fabric workspace ID

  • reportType: Type of report (performance, usage, health, summary)

🚀 Livy API Integration (Sessions & Batch Jobs)

Session Management

• Tool: create-livy-session

• Description: Create a new Livy session for interactive Spark/SQL execution

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • sessionConfig: Optional session configuration

• Tool: get-livy-session

• Description: Get details of a Livy session

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • sessionId: Livy session ID

• Tool: list-livy-sessions

• Description: List all Livy sessions in a lakehouse

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

• Tool: delete-livy-session

• Description: Delete a Livy session

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • sessionId: Livy session ID

Statement Execution

• Tool: execute-livy-statement

• Description: Execute SQL or Spark statements in a Livy session

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • sessionId: Livy session ID

  • code: SQL or Spark code to execute

  • kind: Statement type (sql, spark, etc.)

• Tool: get-livy-statement

• Description: Get status and results of a Livy statement

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • sessionId: Livy session ID

  • statementId: Statement ID

Batch Job Management

• Tool: create-livy-batch

• Description: Create a new Livy batch job for long-running operations

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • batchConfig: Batch job configuration

• Tool: get-livy-batch

• Description: Get details of a Livy batch job

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • batchId: Batch job ID

• Tool: list-livy-batches

• Description: List all Livy batch jobs in a lakehouse

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

• Tool: delete-livy-batch

• Description: Delete a Livy batch job

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Microsoft Fabric lakehouse ID

  • batchId: Batch job ID

📊 Spark Application Monitoring

Workspace-Level Monitoring

• Tool: get-workspace-spark-applications

• Description: Get all Spark applications in a Microsoft Fabric workspace

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • continuationToken: Optional token for pagination

Item-Specific Monitoring

• Tool: get-notebook-spark-applications

• Description: Get all Spark applications for a specific notebook

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • notebookId: Notebook ID

  • continuationToken: Optional token for pagination

• Tool: get-lakehouse-spark-applications

• Description: Get all Spark applications for a specific lakehouse

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • lakehouseId: Lakehouse ID

  • continuationToken: Optional token for pagination

• Tool: get-spark-job-definition-applications

• Description: Get all Spark applications for a specific Spark Job Definition

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • sparkJobDefinitionId: Spark Job Definition ID

  • continuationToken: Optional token for pagination

Application Management

• Tool: get-spark-application-details

• Description: Get detailed information about a specific Spark application

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • livyId: Livy session ID

• Tool: cancel-spark-application

• Description: Cancel a running Spark application

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • livyId: Livy session ID

Monitoring Dashboard

• Tool: get-spark-monitoring-dashboard

• Description: Generate a comprehensive monitoring dashboard with analytics

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

📓 Notebook Management

The MCP server provides comprehensive notebook management capabilities with predefined templates and custom notebook support.

Create Notebook from Template

• Tool: create-fabric-notebook

• Description: Create new Fabric notebooks from predefined templates or custom definitions

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • displayName: Display name for the new notebook

  • template: Template type (blank, sales_analysis, nyc_taxi_analysis, data_exploration, machine_learning, custom)

  • customNotebook: Custom notebook definition (required if template is 'custom')

  • environmentId: Optional environment ID to attach

  • lakehouseId: Optional default lakehouse ID

  • lakehouseName: Optional default lakehouse name

Available Templates:

• blank: Basic notebook with minimal setup

• sales_analysis: Comprehensive sales data analysis with sample dataset

• nyc_taxi_analysis: NYC taxi trip data analysis with sample dataset

• data_exploration: Structured data exploration template

• machine_learning: Complete ML workflow template

• custom: Use your own notebook definition

Get Notebook Definition

• Tool: get-fabric-notebook-definition

• Description: Retrieve the notebook definition (cells, metadata) from an existing Fabric notebook

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • notebookId: ID of the notebook to retrieve

  • format: Format to return (ipynb or fabricGitSource)

Update Notebook Definition

• Tool: update-fabric-notebook-definition

• Description: Update the notebook definition (cells, metadata) of an existing Fabric notebook

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • notebookId: ID of the notebook to update

  • notebookDefinition: Updated notebook definition object

Execute Notebook

• Tool: run-fabric-notebook

• Description: Execute a Fabric notebook on-demand with optional parameters and configuration

• Parameters:

  • bearerToken: Microsoft Fabric bearer token

  • workspaceId: Microsoft Fabric workspace ID

  • notebookId: ID of the notebook to run

  • parameters: Optional notebook parameters (key-value pairs with types)

  • configuration: Optional execution configuration (environment, lakehouse, pools, etc.)

Features:

• 📓 Base64 encoded notebook payload support

• 🔧 Comprehensive metadata management

• 🌐 Environment and lakehouse integration

• 🎛️ Parameterized notebook execution

• ⚡ Spark configuration support

• 🔤 Support for multiple programming languages (Python, Scala, SQL, R)

🚀 Quick Start

🎯 Installation Methods

Choose your preferred installation method:

Option 1: Python Package (PyPI) ⭐ Recommended

Option 2: NPM Package

Option 3: Universal Installation Script

For automated setup with environment configuration:

Unix/Linux/macOS:

Windows (PowerShell):

Option 4: Docker

📖 See

Option 5: From Source (Development)

⚙️ Configuration

Set up your environment variables:

🔧 Claude Desktop Setup

Add to your Claude Desktop config:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

For PyPI Installation:

For NPM Installation:

For Source Installation:

🚀 Start Using

Restart Claude Desktop and try these queries:

• "List all workspaces I have access to"

• "Find workspace named 'Analytics'"

• "List all items in my Fabric workspace [your-workspace-id]"

• "Create a new lakehouse called 'Analytics Hub'"

• "Show me all running Spark applications"

• "Execute this SQL query: SELECT * FROM my_table LIMIT 10"

🧪 Development & Testing

Running the Server

Testing Livy API Integration

For comprehensive testing of Spark functionality, install Python dependencies:

Available Test Scripts:

• livy_api_test.ipynb - Interactive notebook for step-by-step testing

• comprehensive_livy_test.py - Full-featured test with error handling

• spark_monitoring_test.py - Spark application monitoring tests

• mcp_spark_monitoring_demo.py - MCP server integration demo

Claude Desktop Integration

Add this configuration to your Claude Desktop config file:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

🎉 You're ready! Restart Claude Desktop and start asking questions about your Microsoft Fabric data!

Livy API Testing Setup

For testing the Livy API functionality, additional Python dependencies are required:

Available Test Scripts:

• livy_api_test.ipynb - Interactive Jupyter notebook for step-by-step testing

• comprehensive_livy_test.py - Full-featured test with error handling

• simple_livy_test.py - Simple test following example patterns

• livy_batch_test.py - Batch job testing capabilities

• spark_monitoring_test.py - Spark application monitoring tests

Usage

Running the Server

Development Mode

Testing with Claude Desktop

Add the following configuration to your Claude Desktop config file:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

💬 Example Queries

Once connected to Claude Desktop, you can ask natural language questions like:

CRUD Operations:

• "List all Lakehouses in my workspace"

• "Create a new Notebook called 'Data Analysis'"

• "Update the description of my lakehouse"

• "Delete the test notebook from my workspace"

Notebook Management:

• "Create a sales analysis notebook with sample data"

• "Generate a new NYC taxi analysis notebook"

• "Create a machine learning notebook template"

• "Get the definition of my existing notebook"

• "Run my notebook with specific parameters"

• "Update my notebook with new cells"

Data Operations:

• "Query the sales dataset to get total revenue by region"

• "Execute my analytics notebook with today's date"

Analytics:

• "Get performance metrics for the last 24 hours"

• "Analyze my data model and provide optimization recommendations"

• "Generate a usage report for my workspace"

Livy API Operations:

• "Create a Livy session for interactive Spark analysis"

• "Execute SQL query 'SELECT * FROM my_table LIMIT 10'"

• "Run Spark code to show all tables"

• "Monitor my batch job progress"

Spark Application Monitoring:

• "Show me all Spark applications in my workspace"

• "What's the status of my notebook Spark jobs?"

• "Generate a comprehensive Spark monitoring dashboard"

• "Show me recent failed applications"

• "Cancel the problematic Spark application"

Capacity Management:

• "List all Fabric capacities I can use"

• "Assign workspace 1234abcd-abcd-1234-abcd-123456789000 to capacity f9998888-7777-6666-5555-444433332222"

• "Show all workspaces in capacity f9998888-7777-6666-5555-444433332222"

• "Unassign workspace 1234abcd-abcd-1234-abcd-123456789000 from its capacity"

🧩 Capacity Management Tools

Manage Microsoft Fabric capacity assignments directly from your AI assistant. These tools let you inspect available capacities, attach/detach workspaces, and audit capacity usage.

Available Tools

• fabric_list_capacities – Enumerate all capacities you can access (ID, SKU, region, state)

• fabric_assign_workspace_to_capacity – Attach a workspace to a dedicated capacity

• fabric_unassign_workspace_from_capacity – Return a workspace to shared capacity

• fabric_list_capacity_workspaces – List all workspaces currently hosted on a given capacity

Notes

• If authentication fails or you're in simulation mode, capacity responses are simulated.

• Real capacity operations require appropriate Fabric / Power BI admin permissions.

• You can provide a bearer token per call (bearerToken field) or rely on global auth.

Minimal Parameter Reference

❗ Troubleshooting

JSON Parse Errors (e.g., Unexpected token 'P')

If Claude Desktop or another MCP client reports an error like:

This is the EXACT issue reported in GitHub issue where the user saw "Unexpected token 'P', 'Please set'..." errors.

This almost always means something wrote plain text to STDOUT (which must contain ONLY JSON-RPC frames). Common causes:

1. Added console.log debug statements in server code

2. A dependency emitting warnings to STDOUT

3. Early logging before transport initialization

Fixes Implemented in This Server

• ✅ A startup guard now redirects console.log / console.info to STDERR automatically

• ✅ Debug output has been consolidated behind the DEBUG_MCP_RUN=1 flag

• ✅ All diagnostic messages go to STDERR, keeping STDOUT clean for JSON-RPC protocol

What You Can Do

• Avoid adding raw console.log statements—prefer console.error (goes to STDERR)

• If you must allow stdout logging temporarily (NOT recommended), set:

  • ALLOW_UNSAFE_STDOUT=true

  • Remove after debugging

• Regenerate the build (npm run build) after changes to ensure compiled output matches source

Capacity Tools Not Appearing?

If the capacity tools don't show up when the client lists tools:

1. Ensure you rebuilt after pulling changes: npm run build

2. Confirm you're launching the server from build/index.js and not an older snapshot

3. Verify no MCP client-side allowlist is filtering tool names

4. Run a quick enumeration test: ask the assistant "List all available tools"

5. If still missing, delete the build/ folder and rebuild to clear stale artifacts

Authentication Edge Cases

• Azure CLI auth may fail silently without an active az login session

• Bearer tokens expire (~1 hour); refresh if operations suddenly fail

• For local testing: falling back to simulation still lets you prototype tool flows

Getting Verbose Diagnostics

Set the following (sent to STDERR, safe for MCP framing):

Optionally add structured auth tracing:

Need a new troubleshooting topic? Open an issue or PR so others benefit from the resolution.

🔐 Authentication

This MCP server supports multiple authentication methods powered by Microsoft Authentication Library (MSAL):

🤖 For Claude Desktop: Use Bearer Token Authentication (Method #1) for the best experience and compatibility.

🔧 Claude Desktop Fix: Recent updates prevent authentication timeouts by prioritizing bearer tokens and adding timeout protection for interactive authentication flows.

🎫 1. Bearer Token Authentication (Recommended for Claude Desktop)

Perfect for AI assistants and interactive usage:

For Claude Desktop:

• Visit Power BI Embed Setup

• Generate a bearer token for your workspace

• Add to your claude_desktop_config.json

• No timeout issues - bypasses interactive authentication entirely

For Testing:

🤖 2. Service Principal Authentication (Recommended for Production)

Use Azure AD application credentials:

• Client ID (Application ID)

• Client Secret

• Tenant ID (Directory ID)

Environment Variables Setup:

Claude Desktop Configuration:

📱 3. Device Code Authentication

Sign in with browser on another device (great for headless environments):

🌐 4. Interactive Authentication

Automatic browser-based authentication:

🔧 5. Azure CLI Authentication ⭐ (Recommended for Local Development)

Use your existing Azure CLI login for seamless local testing:

Prerequisites:

1. Install Azure CLI: winget install Microsoft.AzureCLI (Windows) or Download

2. Login to Azure: az login

3. Set active subscription: az account set --subscription "your-subscription-name"

Benefits:

• ✅ Zero Configuration - Uses your existing Azure login

• ✅ Instant Setup - No app registration or client secrets needed

• ✅ Multi-Account Support - Switch Azure accounts easily

• ✅ Perfect for Development - Seamless local testing experience

Quick Test:

💡 Pro Tip: Azure CLI authentication is perfect for developers who want to quickly test the MCP server without complex Azure AD app setup. Just az login and you're ready to go!

🔧 Complete Authentication Setup

📚 Detailed Guides:

• Authentication Setup Guide - Complete Azure AD setup

• Claude Desktop Config Examples - Ready-to-use configurations

🔍 Authentication Testing

Check your authentication status:

🔒 Security Best Practices

• Never commit authentication tokens to version control

• Use Service Principal authentication for production deployments

• Device Code flow is perfect for CI/CD and headless environments

• Interactive authentication is ideal for development and testing

• All tokens are automatically validated and include expiration checking

Note: The MCP server seamlessly handles token validation and provides clear error messages for authentication issues.

☸️ Azure Kubernetes Service (AKS) Deployment

Deploy the MCP server as a scalable service on Azure Kubernetes Service for enterprise production use.

🚀 Quick AKS Deployment

Prerequisites

• Azure CLI installed and configured

• Docker installed

• kubectl installed

• Azure subscription with AKS permissions

1. Build and Push Docker Image

2. Deploy to AKS

3. Access the MCP Server

Once deployed, your MCP server will be available at:

🏗️ Architecture Overview

The AKS deployment includes:

• Horizontal Pod Autoscaler (3-10 pods based on CPU/memory)

• Azure Load Balancer for high availability

• SSL/TLS termination with Azure Application Gateway

• ConfigMaps for environment configuration

• Secrets for secure credential storage

• Health checks and readiness probes

• Resource limits and quality of service guarantees

📁 Deployment Files

All Kubernetes manifests are located in the /k8s directory:

• namespace.yaml - Dedicated namespace

• deployment.yaml - Application deployment with scaling

• service.yaml - Load balancer service

• ingress.yaml - External access and SSL

• configmap.yaml - Configuration management

• secret.yaml - Secure credential storage

• hpa.yaml - Horizontal Pod Autoscaler

🔧 Configuration

Configure the deployment by setting these environment variables:

🔐 Production Security

The AKS deployment includes enterprise-grade security:

• Non-root container execution

• Read-only root filesystem

• Secret management via Azure Key Vault integration

• Network policies for traffic isolation

• RBAC with minimal required permissions

• Pod security standards enforcement

📊 Monitoring & Scaling

• Azure Monitor integration for logs and metrics

• Application Insights for performance monitoring

• Prometheus metrics endpoint for custom monitoring

• Auto-scaling based on CPU (70%) and memory (80%) thresholds

• Health checks for automatic pod restart

🔄 CI/CD Integration

The deployment scripts support:

• Azure DevOps pipelines

• GitHub Actions workflows

• Automated testing before deployment

• Blue-green deployments for zero downtime

• Rollback capabilities for quick recovery

📚 Detailed Guide: See AKS_DEPLOYMENT.md for complete setup instructions.

🌐 Azure Model Context Protocol Server (Preview)

Microsoft Azure now offers a preview service for hosting MCP servers natively. This eliminates the need for custom infrastructure management.

🚀 Azure MCP Server Deployment

Prerequisites

• Azure subscription with MCP preview access

• Azure CLI with MCP extensions

Deploy to Azure MCP Service

Configure Authentication

Access Your MCP Server

🔧 Azure MCP Server Features

• Automatic scaling based on usage

• Built-in monitoring and logging

• Integrated security with Azure AD

• Zero infrastructure management

• Global CDN for low latency

• Automatic SSL/TLS certificates

💰 Cost Optimization

Azure MCP Server offers:

• Pay-per-request pricing model

• Automatic hibernation during idle periods

• Resource sharing across multiple clients

• No minimum infrastructure costs

📚 Learn More: Azure MCP Server Documentation

Note: Azure MCP Server is currently in preview. Check Azure Preview Terms for service availability and limitations.

🏗️ Architecture

This MCP server is built with:

• TypeScript for type-safe development

• MCP SDK for Model Context Protocol implementation

• Zod for schema validation and input sanitization

• Node.js runtime environment

⚙️ Configuration

The server uses the following configuration files:

• tsconfig.json - TypeScript compiler configuration

• package.json - Node.js package configuration

• .vscode/mcp.json - MCP server configuration for VS Code

🔧 Development

Project Structure

Adding New Tools

To add new tools to the server:

1. Define the input schema using Zod

2. Implement the tool using server.tool()

3. Add error handling and validation

4. Update documentation

API Integration

This server includes:

✅ Production Ready:

• Full Microsoft Fabric Livy API integration

• Spark session lifecycle management

• Statement execution with SQL and Spark support

• Batch job management for long-running operations

• Comprehensive error handling and retry logic

• Real-time polling and result retrieval

🧪 Demonstration Features:

• CRUD operations (configurable for real APIs)

• Analytics and metrics (extensible framework)

• Data model analysis (template implementation)

🧪 Testing

🚀 End-to-End Testing

The MCP server includes comprehensive end-to-end testing that creates real workspaces, items, and jobs to validate complete functionality using Azure CLI authentication.

Quick Setup for E2E Testing

What the E2E Test Does

The end-to-end test creates a complete workflow in your Microsoft Fabric tenant:

1. 🔐 Validates Azure CLI Authentication - Uses your existing az login session

2. 🏗️ Creates a Test Workspace - New workspace with unique naming

3. ⚡ Attaches to Capacity - Links workspace to your Fabric capacity (optional)

4. 📓 Creates Notebooks & Lakehouses - Test items for validation

5. 🏃 Runs Real Jobs - Executes notebook with actual Spark code

6. 📊 Monitors Execution - Tracks job status and completion

7. 🧹 Cleans Up Resources - Removes all created test resources

E2E Test Configuration

The setup script creates a .env.e2e configuration file:

E2E Test Features

• ✅ Real Resource Creation - Creates actual Fabric workspaces and items

• ✅ Azure CLI Integration - Uses your existing Azure authentication

• ✅ Capacity Assignment - Tests workspace-to-capacity attachment

• ✅ Job Execution - Runs real Spark jobs and monitors completion

• ✅ Automatic Cleanup - Removes all test resources automatically

• ✅ Comprehensive Logging - Detailed logging of all operations

• ✅ Error Handling - Robust error handling and recovery

Prerequisites for E2E Testing

1. Azure CLI installed and logged in:

   az login

2. Microsoft Fabric Access with permissions to:

   • Create workspaces

   • Create notebooks and lakehouses

   • Run Spark jobs

   • (Optional) Assign workspaces to capacity

3. Fabric Capacity (optional but recommended):

   • Set FABRIC_CAPACITY_ID in .env.e2e for capacity testing

   • Without capacity, workspace will use shared capacity

Running E2E Tests

E2E Test Output

The test provides comprehensive output including:

⚠️ Important Notes for E2E Testing

• Creates Real Resources: The test creates actual workspaces and items in your Fabric tenant

• Requires Permissions: Ensure you have necessary Fabric permissions

• Uses Capacity: Jobs may consume capacity units if using dedicated capacity

• Automatic Cleanup: All resources are automatically deleted after testing

• Network Dependent: Requires stable internet connection for API calls

🧪 Unit & Integration Testing

Prerequisites

Available Test Scripts

• livy_api_test.ipynb - Interactive Jupyter notebook for step-by-step testing

• comprehensive_livy_test.py - Full-featured test with error handling

• simple_livy_test.py - Simple test following example patterns

• livy_batch_test.py - Batch job testing capabilities

• spark_monitoring_test.py - Spark application monitoring tests

Quick Testing

1. Interactive Testing:

   jupyter notebook livy_api_test.ipynb

2. Command Line Testing:

   python simple_livy_test.py python spark_monitoring_test.py

3. Comprehensive Testing:

🤝 Contributing

We welcome contributions! Here's how to get started:

1. Fork the repository

2. Create a feature branch (git checkout -b feature/amazing-feature)

3. Make your changes and add tests if applicable

4. Commit your changes (git commit -m 'Add amazing feature')

5. Push to the branch (git push origin feature/amazing-feature)

6. Open a Pull Request

Development Guidelines

• Follow TypeScript best practices

• Add JSDoc comments for new functions

• Update tests for any new functionality

• Update documentation as needed

• See CONTRIBUTING.md for detailed guidelines

🔒 Security

• Never commit authentication tokens to version control

• Use environment variables for sensitive configuration

• Follow Microsoft Fabric security best practices

• Report security issues privately via GitHub security advisories

• See SECURITY.md for our full security policy

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

For issues and questions:

• 📖 Check the MCP documentation

• 📚 Review Microsoft Fabric API documentation

• 🐛 Open an issue in this repository

• 💬 Join the community discussions

Acknowledgments

• Microsoft Fabric Analytics team for the comprehensive data platform and analytics capabilities

• Microsoft Fabric Platform teams for the robust API platform and infrastructure

• Bogdan Crivat and Chris Finlan for the inspiring brainstorming conversation that gave me the idea to open-source this project

• Anthropic for the Model Context Protocol specification

This project began as my weekend hack project exploring AI integration with Microsoft Fabric. During a casual conversation with Chris and Bogdan about making AI tooling more accessible. What started as a personal experiment over a weekend is now available for everyone to build upon.