MCP Server¶

Use fab-bundle as an MCP (Model Context Protocol) server in GitHub Copilot, Claude Code, or any MCP-compatible client. This lets you manage Fabric workspaces conversationally.

Install¶

pip install fabric-automation-bundles[mcp]

Configure¶

GitHub Copilot¶

Add to your project's .vscode/mcp.json:

{
  "servers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    }
  }
}

Or add to your VS Code user settings (settings.json) under "mcp.servers" to use across all projects.

Claude Code¶

Add to your project's .claude/settings.json:

{
  "mcpServers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    }
  }
}

Or add globally to ~/.claude/settings.json to use across all projects.

Claude Desktop¶

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    }
  }
}

Prerequisites¶

Before using the MCP server, make sure:

Azure CLI authenticated: az login
fabric.yml exists in your project directory
Fabric capacity is active and accessible

Run fab-bundle doctor to verify everything is configured correctly.

Available Tools¶

Tool	Description	Example prompt
`fab_validate`	Validate fabric.yml	"Validate my Fabric bundle"
`fab_plan`	Preview deployment changes	"What would change if I deploy to dev?"
`fab_deploy`	Deploy to a target	"Deploy to dev"
`fab_destroy`	Tear down resources	"Destroy the test environment"
`fab_status`	Show deployed resources	"What's deployed in prod?"
`fab_drift`	Detect out-of-band changes	"Check for drift in staging"
`fab_run`	Run a notebook or pipeline	"Run the ingest_to_bronze notebook in dev"
`fab_history`	Show deployment history	"Show me recent deployments"
`fab_doctor`	Diagnose issues	"Check if my Fabric setup is working"
`fab_list_templates`	List templates	"What templates are available?"
`fab_list_workspaces`	List workspaces	"Show me all Fabric workspaces"
`fab_list_capacities`	List capacities	"What Fabric capacities do I have?"

Example Conversations¶

"What's in my Fabric workspace?"

The AI calls fab_status and responds:

"Your dev workspace has 12 items deployed: 3 lakehouses (bronze, silver, gold), 3 notebooks (ingest_to_bronze, transform_to_silver, aggregate_to_gold), 2 pipelines (daily_etl, weekly_refresh), 1 Spark environment, 1 warehouse, 1 SQL analytics endpoint, and 1 data agent. Last deployment was 3 hours ago. No drift detected."

"Deploy my project to dev"

The AI first calls fab_validate to check your bundle, then fab_plan to preview changes:

"Your fabric.yml is valid. Here's what will happen: 3 lakehouses will be created, 3 notebooks uploaded, and 2 pipelines configured with schedules. Total: 8 new resources. No existing resources will be modified. Should I proceed with the deployment?"

After you confirm, it calls fab_deploy and reports:

"Deployed 8 resources to sales-analytics-dev in 45 seconds. All items created successfully. You can open the workspace at https://app.fabric.microsoft.com/..."

"Something changed in prod that I didn't deploy"

The AI calls fab_drift and responds:

"I found drift in 2 items. A notebook 'manual_fix' was added directly in the portal (not in your fabric.yml). And the 'daily_etl' pipeline schedule was changed from 6:00 AM to 8:00 AM. Would you like me to update your fabric.yml to match, or redeploy to revert the changes?"

"Run the ETL pipeline"

The AI calls fab_run and responds:

"I've submitted the 'daily_etl' pipeline in the dev workspace. Job ID: a3f8c... You can monitor progress in the Fabric portal under Monitoring Hub. I'll check back -- want me to poll the status?"

"Set up a new Fabric project"

The AI calls fab_list_capacities and fab_list_templates, then responds:

"You have 2 capacities available: 'Team-Dev' (F8, West US 2) and 'Prod' (F16, East US). I found 4 templates: medallion (bronze/silver/gold lakehouse), starter (single lakehouse + notebook), warehouse (SQL-first with views), and custom (empty scaffold). Which template and capacity would you like to use?"

Setting up both MCP servers¶

For the best experience, use fab-bundle-mcp alongside Microsoft's Fabric MCP server. fab-bundle-mcp lets the AI act on your workspace (deploy, plan, status), while Microsoft's Fabric MCP gives the AI deep knowledge of Fabric APIs and best practices.

GitHub Copilot (`.vscode/mcp.json`)¶

{
  "servers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    },
    "fabric": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-server-fetch", "https://github.com/microsoft/mcp"]
    }
  }
}

Check Microsoft's Fabric MCP repo for the latest install command -- the args above are illustrative.

Claude Code (`.claude/settings.json`)¶

{
  "mcpServers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    },
    "fabric": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-server-fetch", "https://github.com/microsoft/mcp"]
    }
  }
}

Claude Desktop (`claude_desktop_config.json`)¶

{
  "mcpServers": {
    "fab-bundle": {
      "command": "fab-bundle-mcp"
    },
    "fabric": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-server-fetch", "https://github.com/microsoft/mcp"]
    }
  }
}

What each server provides¶

MCP Server	Purpose	Example
fab-bundle-mcp	Manage your Fabric project: deploy, plan, status, run, drift, destroy	"Deploy to dev", "Check for drift"
Microsoft Fabric MCP	Fabric API docs, best practices, item schemas	"How do I configure a pipeline trigger?", "What Spark versions does Fabric support?"

Together, the AI can both understand Fabric and act on your workspace.

Troubleshooting¶

"fab-bundle-mcp: command not found" - Make sure you installed with pip install fabric-automation-bundles[mcp] - Check that the pip scripts directory is in your PATH

"Authentication error" - Run az login in your terminal first - Or set AZURE_TENANT_ID, AZURE_CLIENT_ID, AZURE_CLIENT_SECRET environment variables

Tools not showing up - Restart your IDE after adding the MCP configuration - Check the MCP server logs for errors