Model Context Protocol Power BI MCP Server — Complete Guide

What Is the Model Context Protocol and Why Does It Matter for Power BI?

The Model Context Protocol (MCP) is an open standard that defines how AI assistants interact with external tools and data sources in a structured, secure way. It establishes a common interface so that any MCP-compatible AI client — GitHub Copilot, Claude, custom agents — can connect to any MCP-compatible server without custom integration code for each combination.

Before MCP, connecting an AI assistant to Power BI required custom REST API calls, manual DAX construction, and separate authentication handling per application. With Power BI MCP servers, any AI client that supports MCP discovers and invokes Power BI tools through a standardized protocol — no custom integration code needed per client.

The Three MCP Components

In a typical setup: VS Code is the host, GitHub Copilot is the client, and the Power BI MCP server provides the tools. When you ask Copilot “what was Q2 revenue by product?”, Copilot uses the MCP server to query your Power BI semantic model and return the answer — all within VS Code.

Two Power BI MCP Servers — Remote vs Local

Microsoft provides two distinct MCP server implementations for Power BI, each serving a different persona and workflow. Choosing the wrong one for a use case is the most common setup mistake.

Remote Power BI MCP Server — Query Data with Natural Language

The remote Power BI MCP server is a hosted endpoint that translates natural language prompts into DAX queries, executes them against Power BI semantic models, and returns results to the AI agent. It uses Copilot’s intelligence for DAX generation by default — consuming Copilot capacity. If you prefer to use your MCP client’s own LLM for DAX generation, disable the Copilot tool in your MCP client configuration.

What the Remote Server Can Do

• Execute natural language queries against Power BI semantic models
• Generate and run DAX queries from conversational prompts
• Retrieve semantic model metadata — tables, columns, measures, relationships
• Retrieve Power BI report schema — pages, visuals, filters, workspace info
• Support any MCP-compatible LLM — not limited to Microsoft models
• Respect existing Power BI permissions for user identity authentication

Managing Semantic Model IDs

Every tool call to the remote MCP server requires a semantic model ID. Rather than requiring users to provide this in every session, store model IDs in a accessible location:

# semantic-model-ids.json (VS Code workspace file)
{
  "models": {
    "sales_analytics": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
    "finance_summary": "yyyyyyyy-yyyy-yyyy-yyyy-yyyyyyyyyyyy",
    "hr_dashboard":    "zzzzzzzz-zzzz-zzzz-zzzz-zzzzzzzzzzzz"
  }
}

Retrieve a semantic model ID from the Power BI service URL when viewing the model, or programmatically via the Power BI REST API. For multi-model scenarios, maintain a catalog mapping friendly names to model IDs in your agent’s configuration files.

Local Power BI MCP Server — Build and Modify Semantic Models

The local Power BI MCP server is an open-source project published by Microsoft at github.com/microsoft/powerbi-modeling-mcp. It runs on your machine, connects to your semantic model using your existing credentials, and gives AI agents comprehensive modeling capabilities — from simple property updates to complex agentic workflows involving TMDL files and bulk operations.

What the Local Server Can Do

Recommended AI Models for the Local Server

Microsoft recommends deep-reasoning models such as GPT-5 or Claude Sonnet 4.5 for best results. These handle complex DAX generation and schema reasoning more reliably than standard models. Be aware of token costs on large models — a medium-sized model.bim file (~26,000 lines JSON) requires approximately 210,000 tokens, which may exceed standard context windows and incurs significant cost per session. For large models, scope prompts to specific tables or measure groups rather than loading the entire schema.

Setting Up the Power BI MCP Server

Remote MCP Server Setup

1. Enable the tenant settingA Power BI administrator must enable “Users can use the Power BI Model Context Protocol server endpoint (preview)” in the Power BI admin portal. Without this, all connection attempts fail regardless of user permissions.
2. Add the remote server to your MCP clientIn VS Code with GitHub Copilot, add the Power BI remote MCP server endpoint to your MCP client configuration. The server URL is available from the Microsoft Learn documentation for the remote server.
3. AuthenticateUse user identity (delegated) authentication for individual use. For automated agents, configure Service Principal authentication — but read the RLS warning in Section 08 first.
4. Store semantic model IDsCreate a semantic-model-ids.json file in your workspace with friendly name to model ID mappings. Each tool call requires a model ID — store them where your agent can read them automatically.
5. Test with a simple queryAsk your AI client “using the [model name] semantic model, what are the available tables?” This calls get_semantic_model_info and confirms the connection is working before running DAX queries.

Local MCP Server Setup

# Install the local Power BI MCP server
npm install -g @microsoft/powerbi-modeling-mcp

# Add to VS Code MCP configuration (.vscode/mcp.json)
{
  "powerbi-modeling": {
    "command": "powerbi-modeling-mcp",
    "args": ["--stdio"]
  }
}

# Connect to Power BI Desktop (auto-detects running instances)
# Prompt: "Find and connect to my Sales Model Power BI Desktop file"

# Connect to a Fabric workspace model
# Prompt: "Connect to the Sales Analytics model in the Production workspace"

# With Service Principal authentication
powerbi-modeling-mcp --stdio   --authmode serviceprincipal   --tenantid YOUR_TENANT_ID   --clientid YOUR_CLIENT_ID

Power BI MCP Server Tools Reference

Remote Server Tools

Local Server Key Operations

Authentication & Security Model

Both Power BI MCP servers handle credentials through Microsoft’s official authentication infrastructure. Neither server stores or manages tokens directly — all credential handling goes through the Azure Identity SDK.

Microsoft Security Guidance for Power BI MCP

• Enable Entra ID authentication — do not expose MCP endpoints without identity verification
• Use secure token management — credentials through Azure Identity SDK only
• Apply network isolation — restrict which networks can reach MCP server endpoints
• Apply least-privilege RBAC roles before deployment — Fabric RBAC permissions govern what operations MCP clients can invoke
• Review your LLM provider’s data handling policies — model metadata and query results are forwarded to the LLM as context

RLS Warning — Service Principal Authentication Does Not Enforce Row Level Security

This is the most critical security consideration for the remote Power BI MCP server and must be understood before any production deployment.

When RLS Is Enforced

Row Level Security is enforced normally when using user identity (delegated) authentication. The remote MCP server executes queries in the context of the signed-in user, and all RLS filters that apply to that user apply to every query the agent makes on their behalf. For agents that serve individual named users, delegated authentication is the correct and secure choice.

Production Decision Framework

DAX Generation — Getting the Best Results

DAX generation quality from the Power BI MCP server depends significantly on how well the semantic model is prepared for AI and how queries are framed. Poor model metadata produces poor DAX — regardless of which LLM is generating it.

Prepare Your Semantic Model for MCP

1. Add table and column descriptionsDescriptions in the semantic model are returned by get_semantic_model_info and directly inform DAX generation. A measure described as “Net Revenue = Gross Revenue minus approved refunds” produces far more accurate DAX than one with no description.
2. Configure AI-optimized metadataThe remote server retrieves Copilot tooling metadata when configured — this provides additional context about model structure and guides the agent toward the correct data. Configure this in Power BI Desktop under Model settings.
3. Call get_semantic_model_info before execute_queryGround DAX generation in the actual model structure before asking a question. Agents that skip the schema retrieval step frequently generate DAX that references non-existent tables or uses incorrect column names.
4. Use get_report_info to understand filter contextReports reveal how the semantic model is used in practice — which filters are applied, which visuals are standard, what the intended context is. This context significantly improves DAX generation accuracy for questions related to an existing report.

Power BI MCP Server — Current Limitations

FAQ — Model Context Protocol Power BI MCP Server

Official Resources — Power BI MCP Server Documentation