Back office ops · Production

Authoring Microsoft Fabric artifacts with natural language via MCP and Agent Skills

The problem

Creating Fabric artifacts such as DAX measures, lakehouses, and delta tables previously required deep platform knowledge, precise scripting, and sequences of manual UI interactions.

Workflow diagram · grounded in source

Natural language instruction

trigger

“You describe your intent using natural language: "In Contoso SemanticModel, create a DAX measure in the Sales table that calculates the total count of rows."”

AI resolves workspace context

ai_action

“Copilot understands: - Which workspace you're targeting - Which semantic model is in context - Which table you're referencing - What kind of aggregation you want”

MCP API execution

integration

“Fabric MCP selects the appropriate tooling, connects to the semantic model, generates valid DAX, and executes the operation directly”

Agent plans and executes via CLI

ai_action

“the agent: - Interprets the business intent - Resolves workspace and model identity - Generates the DAX expression - Uses the Fabric CLI to apply the change - Explains exactly what was created”

Proactive error correction

validation

“Detects that the file is actually a CSV, not XLSX - Adjusts the ingestion approach automatically”

Fabric artifacts delivered

output

“The file is uploaded to the lakehouse and it even creates a notebook with some code that it needed to create the delta table”

Reported outcome

Both Fabric MCP and Agent Skills deliver a conversational authoring experience where natural language instructions produce Fabric artifacts without platform-specific syntax or manual steps; the agent also adapts proactively to errors such as incorrect file extensions.

Reported metrics

Manual developer steps eliminatednever touch DAX syntax, don't navigate the Fabric UI, don't manually deploy or validate

Developer experience qualityinstant, intentional, and safe

Reported stack

Fabric MCPFabric Agent SkillsFabric CLIGitHub CopilotVisual Studio Code

Source

https://medium.com/data-science-at-microsoft/one-experience-two-approaches-authoring-fabric-artifacts-with-mcp-and-agent-skills-bf8af068273f

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Both Fabric MCP and Agent Skills deliver a conversational authoring experience where natural language instructions produce Fabric artifacts without platform-specific syntax or manual steps; the agent also adapts proac…

What tools did this team use?

Fabric MCP, Fabric Agent Skills, Fabric CLI, GitHub Copilot, Visual Studio Code.

What results were reported?

Manual developer steps eliminated: never touch DAX syntax, don't navigate the Fabric UI, don't manually deploy or validate; Developer experience quality: instant, intentional, and safe (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Natural language instruction → AI resolves workspace context → MCP API execution → Agent plans and executes via CLI → Proactive error correction → Fabric artifacts delivered.