How does MCP Engine work?

MCP Engine bridges the gap between AI assistants and Power BI using the Model Context Protocol. It runs locally on your machine, parsing natural language requests, translating them to TOM operations, and executing via ADOMD.NET. Supports Power BI Desktop (Windows) and Power BI Service via XMLA endpoints.

What can MCP Engine do?

MCP Engine can query 15 object types, manage semantic layers, create and modify measures, tables, calculated columns, KPIs, calculation groups, run DAX queries, analyze query performance, manage security roles, and more - all through natural language.

Who is MCP Engine for?

MCP Engine is for Power BI developers, data analysts, BI architects, and anyone who works with Power BI semantic models and wants to accelerate their workflow using AI assistants like Claude or Copilot.

How does it compare to other tools?

Unlike Copilot in Power BI which focuses on report building, MCP Engine focuses on semantic model development. It provides full CRUD operations on the data model, works with any MCP-compatible AI client, and runs entirely locally for privacy.

Yes. MCP Engine runs entirely on your machine. No cloud services, no tracking, no data collection. Your models and data never leave your environment. Pro and Enterprise tiers add PII & numeric masking for additional protection.

Back to home

Frequently Asked Questions

Everything you need to know about MCP Engine.

General Capabilities Getting Started Security & Privacy Troubleshooting

01 — GENERAL

General

Learn what MCP Engine is, who it's for, and how it works.

MCP Engine bridges the gap between AI assistants and Power BI using the Model Context Protocol.

AI CLIENT

Claude / Copilot / Any MCP Client

Natural language requests from any MCP-compatible assistant

MCP

MCP ENGINE

Local Processing

• Parse natural language intent
• Translate to TOM operations
• Execute via ADOMD.NET
• Return structured results

TOM

POWER BI

Semantic Model

Desktop (.pbix) or Service via XMLA endpoint

Model Context Protocol (stdio)

Tabular Object Model (local)

MCP Engine (your machine)

Supports Power BI Desktop (Windows) and Power BI Service via XMLA endpoints (Premium/Fabric capacity)

For Developers

Create measures with natural language, analyze DAX performance

KEY USE CASES

Generate complex DAX measures in seconds

Debug query performance with SE/FE metrics

Explore model schema programmatically

Integrate with CI/CD pipelines

EXAMPLE PROMPT

"Create a Year-over-Year Sales measure with fiscal calendar support"

Feature	MCP Engine	DAX Studio	Tabular Editor	Copilot
AI & Integration
AI-Native Workflow				Generic
MCP Protocol Server
Power BI Automation Tools	15 tools
Query & Analysis
Model Discovery & Connection	Desktop + Service
DAX Query Execution
Query Performance Profiling	Full analysis
Storage & Memory Analysis
Model Management
Schema Management		Read only
Semantic Layer (Measures, KPIs)
DAX & M Expression Formatting		DAX only	DAX only
User-Defined Functions
Security (RLS/OLS) & Perspectives
User Impersonation	Pro
Multi-Language Translation	40+ cultures		Manual
Data Refresh Control
Operations & Safety
Bulk Operations with Dry-Run			Scripting
Read-Only Browse Mode
Data Masking (PII/Numeric)
Governance
Governance & Policy Engine	Pro
Snapshots & Rollback	Pro		Git sync
Audit Logging	Enterprise
Preferences & Rules Memory			BPA rules
Cleanup
Unused Object Detection

Full support Partial Not available

Based on publicly available features as of 2025.

Why MCP Engine?

First AI-Native Power BI Tool

15 MCP tools built from the ground up for AI assistants—not retrofitted

Complete Automation Surface

Query, schema, semantic layer, security, localization, and refresh in one place

Safe by Default

Dry-run mode, read-only browsing, and PII masking protect your models

Enterprise Governance

Policy engine, tamper-evident audit logs, snapshots, and rollback

No. MCP Engine for Power BI is an independent, community-developed tool and is not affiliated with Microsoft.

02 — CAPABILITIES

Capabilities

Explore what MCP Engine can do for your Power BI workflow.

15 Queryable Object Types

Full semantic model coverage

Measures

DAX calculations & aggregations

KPIs

Key performance indicators

Calculation Groups

Reusable calculation logic

UDFs

User-defined functions

Tables

Data tables & fact tables

Columns

Fields & calculated columns

Partitions

M/SQL/DAX/Entity partitions

Relationships

Table connections & cardinality

Hierarchies

Drill-down structures

RLS Roles

Row-level security filters

OLS Roles

Object-level security

Perspectives

Filtered model views

Cultures

40+ regional formats

Translations

Multi-language labels

VertiPaq Stats

Compression & memory analysis

Standout Capabilities

What sets MCP Engine apart

Natural Language Building

Create complex DAX measures with plain English

"Create a Year-over-Year Sales measure"

Bulk Operations

Batch changes with dry-run preview

Rename 50 measures in seconds

Performance Analysis

SE/FE metrics, query plans, VertiPaq optimization

Storage Engine + Formula Engine

Auto-Documentation

Generate descriptions for measures and tables

AI-powered documentation

Multi-Language Support

Translate models automatically to 40+ cultures

Automatic translation

Unified Partition Management

PRO

Manage M, SQL, DAX, and Entity partitions uniformly

All partition types

Snapshots & Rollback

PRO

Preview changes, diff, and safely rollback

Git-like workflow

Policy Engine

PRO

Enforce naming conventions and governance rules

Allow/deny/require-confirm

Safety & Governance

Enterprise-grade protection

Read-Only Mode

Safe exploration without accidental changes. Set via environment variable.

Transaction History

Full audit trail of all operations. Snapshots and rollback support.

PII & Numeric Masking

Automatically mask sensitive data in outputs. Numerical masking included.

Yes, in full mode it exposes authoring tools for creating, updating, and deleting model objects.

Tip: Use MCP_ENGINE_MODE=readonly for non-mutating exploration and analysis only.

Tool visibility depends on:

•Mode — read-only/browse-only hides write tools
•License tier — Free/Pro/Enterprise gate some tools
•Explicit settings — Some tools (e.g., audit) are opt-in

Check manage_license → status for details.

03 — GETTING STARTED

Getting Started

Platform requirements, downloads, and installation guidance.

Power BI Desktop

Windows only (Desktop build)
Windows 10/11 x64
Requires .pbix open in Desktop

Power BI Service (XMLA)

Windows or macOS (XMLA build)
Premium or Fabric capacity
XMLA endpoint enabled

Protocol: MCP over stdio (JSON-RPC 2.0, Content-Length framing). Works with any MCP client that supports stdio transport.

Choose the right build for your platform and use case.

Windows

Power BI Desktop (Local)

Connect to .pbix files opened in Power BI Desktop

Full Build — Read + write tools for full authoring

Read-Only Build — Non-mutating tools only

.mcpb Package — Easiest install for Claude Desktop

macOS

Power BI Service (XMLA Only)

Connect to Power BI Service via XMLA endpoint

XMLA-Only Build — Service connectivity only (no Desktop support on macOS)

Note: Tool visibility depends on runtime mode (MCP_ENGINE_MODE) and license tier. Your MCP client starts the server binary automatically.

View Installation Guide

Defaults are designed to be LLM-friendly. You can lower caps with environment variables:

MCP_ENGINE_MAX_QUERY_ROWS=500

MCP_ENGINE_MAX_PREVIEW_ROWS=500

The server ships 20+ markdown docs in Resources/*.md, also available via MCP resources.

Access via your AI assistant:

"Read the measure authoring guide"

Or use resources/read with URIs like docs://measure-authoring-guide

View all guides

04 — SECURITY & PRIVACY

Security & Privacy

Data handling, safety modes, and privacy features.

MCP Engine runs entirely on your machine. No cloud, no tracking, no data collection. Your models stay yours.

LOCAL_ONLY

No data collection

Your queries and model data never leave your machine

No external uploads

All processing happens locally via stdio

Your models stay yours

Zero telemetry, zero tracking, zero analytics

Safety Modes

read-only

Hides write-capable tools. Safe exploration and analysis only.

browse-only

Leak-minimized mode. Suppresses sensitive fields and restricts operations.

Pro & Enterprise Features

PII & Numeric Masking

Redact sensitive data in outputs

Policy Engine

Allow/deny/confirm rules

Audit Logging

Tamper-evident operation logs

Enterprise tier includes the Admin Policy Bundle feature for centralized policy control.

What it does

• Deploy organization-wide policies via a single JSON file
• Prevent local policy modifications (lockdown mode)
• Automatic protection against misconfiguration

How to enable

Set MCP_ENGINE_POLICY_BUNDLE_PATH to point to your centrally-managed policy file. Requires an active Enterprise license.

Bundle format

{
  "version": "1.0",
  "scope": "global",
  "rules": [
    {
      "id": "block-deletes",
      "action": "deny",
      "tool": "manage_*",
      "message": "Delete operations blocked."
    }
  ]
}

When the bundle is active, all local policy operations are disabled and users follow the centralized rules.

05 — TROUBLESHOOTING

Troubleshooting

Common issues and how to resolve them.

Cause: Power BI Desktop isn't running or no .pbix file is open.

Fix: Open Power BI Desktop and load a .pbix file, then retry the connection.

Cause: Multiple Power BI instances are open and none has been selected yet.

Fix: Use manage_model_connection with operation="list" then operation="select".

Cause: The tool is hidden due to mode or license restrictions.

Fix: Check your configuration:

• Verify MCP_ENGINE_MODE isn't set to read-only/browse-only
• Run manage_license → status to check tier

Cause: Missing or expired credentials for Power BI Service.

Fix: Use manage_model_connection with operation="authenticate" to sign in again.

MCP Engine writes diagnostic logs to stderr (stdout is reserved for MCP protocol messages).

Option 1: MCP Client Logs

Most MCP clients surface server logs from stderr in their developer console or log viewer.

Option 2: Run Directly

Run the binary directly from terminal to see stderr logs in real-time.

Run through this checklist when something isn't working:

Power BI Desktop is running with a .pbix file open

You've run manage_model_connection → list → select

MCP_ENGINE_MODE is not set (or set to your intended mode)

Your license tier supports the tool you're trying to use

For XMLA: you've authenticated and have workspace access

Still have questions?

Check out the documentation or join our community.

Read the Docs View All Features Join the Discussion

Frequently Asked Questions

General

How does MCP Engine work?

Who is MCP Engine for?

For Developers

KEY USE CASES

EXAMPLE PROMPT

How does it compare to other tools?

Why MCP Engine?

First AI-Native Power BI Tool

Complete Automation Surface

Safe by Default

Enterprise Governance

Is this an official Microsoft product?

Capabilities

What can MCP Engine do?

15 Queryable Object Types

Standout Capabilities

Natural Language Building

Bulk Operations

Performance Analysis

Auto-Documentation

Multi-Language Support

Unified Partition Management

Snapshots & Rollback

Policy Engine

Safety & Governance

Read-Only Mode

Transaction History

PII & Numeric Masking

Can it modify my model?

Why is a tool missing from tools/list?

Getting Started

What platforms are supported?

Power BI Desktop

Power BI Service (XMLA)

Which build should I download?

Windows

macOS

How do I cap result sizes?

Where are the built-in manuals?

Security & Privacy

Is my data safe?

No data collection

No external uploads

Your models stay yours

Safety Modes

Pro & Enterprise Features

How do I deploy policies across my organization?

Troubleshooting

"No Power BI Desktop instances found"

"No Power BI model connected"

"Tool not found" in tools/list

"Authentication failed" (XMLA)

How do I view server logs?

Quick troubleshooting checklist

Still have questions?