The release of Power BI MCP Servers represents a structural shift in how AI agents interact with Power BI semantic models moving from the Copilot-within-the-report-pane model to a standardised, open protocol connection that allows any compatible AI agent to query, explore, and modify Power BI data and metadata programmatically. For enterprise data teams that have already integrated AI agents into their development and analytical workflows, this is the mechanism that closes the gap between those agents and the organisation's governed Power BI data assets.

What the Model Context Protocol Is

The Model Context Protocol (MCP) is an open standard, originally developed by Anthropic and now supported across the AI tooling ecosystem, that defines how AI language model agents connect to external data sources, tools, and services in a structured, standardised way. Before MCP, each AI agent that needed to interact with an external service a database, a file system, a BI tool required a bespoke integration: custom API calls, proprietary connectors, or manual copy-paste workflows. MCP provides a common protocol layer that an AI agent can use to discover what capabilities a connected service offers, send requests to invoke those capabilities, and receive structured responses without custom integration code on the AI side.

An MCP server is the component that implements the protocol on the service side. It exposes a defined set of tools callable functions that the AI agent can invoke and resources queryable data that the agent can read. The AI agent connects to the MCP server, discovers its tools and resources, and calls them as needed to complete a task. This separation of protocol from implementation means that any AI agent that speaks MCP can connect to any MCP server, regardless of which AI provider built the agent or which service built the server.

"MCP is to AI agents what REST was to web applications a common protocol that means any agent can connect to any service without either side needing to know the other's internals. Power BI MCP Servers are Power BI's participation in that ecosystem."

Why Power BI Needs MCP Servers

Power BI's existing AI integration - Copilot in reports, the standalone Chat with Your Data experience, Smart Narratives is delivered through Microsoft's own AI infrastructure, tightly coupled to the Power BI Service. These are valuable capabilities for end users, but they are not accessible to AI agents operating outside the Power BI Service interface developer tools, custom agentic workflows, CI/CD pipelines, or third-party AI assistants.

Enterprise data teams increasingly build development and operational workflows around AI agents: GitHub Copilot for code assistance, Claude for documentation and analysis, custom agents for automated data quality checks or pipeline monitoring. Without Power BI MCP Servers, these agents have no structured, governed way to interact with Power BI semantic models. They cannot query a semantic model's measure definitions, generate DAX against a live dataset, explore a model's schema, or retrieve current data values through a governed API. MCP servers provide exactly this a structured, authenticated, permission-controlled interface that lets any MCP-compatible AI agent interact with Power BI as a first-class connected service.

The Two Power BI MCP Servers: Modeling and Remote

Microsoft has published two distinct Power BI MCP Servers, each serving a different interaction pattern between AI agents and Power BI. Understanding the distinction is essential for deciding which server to connect to for a given use case.

The Modeling MCP Server is oriented toward semantic model development workflows. It provides tools that an AI agent can use to read, understand, and modify the structure of a Power BI semantic model measure definitions, table schemas, relationships, calculation groups, and TMDL content. This server is designed for use by developers working on semantic model creation and maintenance, using AI agents to accelerate DAX authoring, model documentation, and structural review.

The Remote MCP Server is oriented toward live data access workflows. It provides tools that an AI agent can use to execute DAX queries against published Power BI datasets in the Power BI Service, retrieve data from connected Fabric Lakehouses and warehouses via the SQL Analytics Endpoint, and interact with Power BI REST API operations. This server is designed for analytical workflows where an AI agent needs to retrieve actual data values, run computations, or orchestrate data operations against the production Power BI estate.

Modeling MCP: Semantic Model Development with AI Agents

The Modeling MCP Server is deployed locally alongside Power BI Desktop or used in conjunction with TMDL-based model development workflows. It connects to a local or workspace-backed semantic model and exposes it to an AI agent through a set of tools covering model schema inspection, measure authoring support, DAX expression generation and validation, and TMDL content read/write operations.

A developer using Claude or GitHub Copilot within VS Code can connect their AI assistant to the Modeling MCP Server and ask questions that the assistant answers by actually querying the semantic model's metadata rather than generating responses from training data alone. Asking "what measures are defined in this model?" returns the actual measures from the connected model. Asking "write a running total measure for Sales Amount" generates DAX that the assistant can validate against the actual measure and table names available in the model, rather than producing a generic template.

For enterprise semantic model development teams, the Modeling MCP Server addresses the documentation and review use cases that traditionally required manual effort: generating a complete measure catalogue with descriptions, reviewing a model's DAX for pattern consistency, identifying measures that reference deprecated columns, or producing TMDL diff summaries for change review processes. These tasks are automatable through an AI agent connected to the Modeling MCP Server in a way that was not possible through the standard Power BI Desktop interface alone.

Remote MCP: Live Fabric Data Access for AI Agents

The Remote MCP Server connects to published datasets in the Power BI Service and to Fabric data sources Lakehouses, Warehouses, and the SQL Analytics Endpoint providing AI agents with the ability to execute live queries and retrieve actual data values. An AI agent connected to the Remote MCP Server can execute a DAX query against a published semantic model, retrieve results as structured data, and use those results as context for further reasoning or reporting.

This capability matters for agentic analytical workflows that go beyond what Copilot's report-level interaction supports. An agent tasked with producing a monthly variance analysis report can use the Remote MCP Server to query the relevant semantic model measures directly, retrieve the current period and prior period values, compute the variance, and generate a narrative summary - all without a human intermediary composing the queries or copying results. The agent has governed, authenticated access to the data through the MCP protocol rather than through screen scraping, CSV exports, or informal API calls.

The Remote MCP Server also exposes Power BI REST API operations, enabling agents to orchestrate service-level operations: triggering dataset refreshes, retrieving refresh history, listing workspaces and datasets, and accessing usage metrics. For agentic monitoring and operations workflows, where an agent is responsible for detecting and escalating dataset refresh failures, for instance, this REST API exposure through the MCP protocol provides a structured, auditable interface for those operations.

AI Clients: Claude, GitHub Copilot, and Custom Agents

Both Power BI MCP Servers are compatible with any AI client that implements the MCP protocol. The clients with the most developed MCP support at the time of the servers' release are Claude Desktop (Anthropic's desktop AI assistant), GitHub Copilot in VS Code and Visual Studio, and custom agents built on frameworks such as LangChain, Semantic Kernel, or AutoGen that have implemented MCP client capability.

For enterprise development teams using GitHub Copilot as their primary AI coding assistant, the Modeling MCP Server integration means that Copilot can provide Power BI-aware suggestions that reference the actual structure of the connected semantic model measure names, table schemas, DAX patterns already present in the model rather than generating generic Power BI advice. For teams using Claude as a document analysis and reasoning tool, the Remote MCP Server integration means that Claude can pull live data from Power BI to ground its analytical responses in current rather than hypothetical values.

Security Architecture: Entra ID, OAuth, and Data Boundary

Both Power BI MCP Servers authenticate using Microsoft Entra ID (formerly Azure Active Directory) through the standard OAuth 2.0 device code flow. When an AI agent first connects to a Power BI MCP Server, it initiates an OAuth flow that prompts the user to authenticate with their Microsoft organisational identity. The resulting access token scopes the agent's permissions to exactly what the authenticated user is permitted to access within the Power BI Service - it cannot access data or metadata that the user's own account cannot access.

This means that all existing Power BI security controls - Row-Level Security in semantic models, object-level security, workspace access controls, and sensitivity labels apply fully to operations performed through the MCP servers. An AI agent querying a dataset through the Remote MCP Server retrieves only the data that the authenticated user's security context permits. There is no elevated privilege, no service account bypass, and no data access that exceeds what the connected identity would have through the standard Power BI Service interface.

Data Boundary and Confidentiality Considerations

When an AI agent uses an MCP server to retrieve data and processes it through an external AI model - Claude, GPT, Gemini, the data leaves the Microsoft trust boundary and enters the AI provider's processing infrastructure. For data classified as confidential or subject to regulatory controls, enterprise governance policies must determine whether MCP-mediated access by external AI agents is permissible, under what conditions, and for which datasets. This is a data classification and information governance decision that should be addressed explicitly before MCP server access is enabled broadly across a Power BI estate.

Modeling MCP vs Remote MCP: Decision Guide

Enterprise Governance Considerations

Enabling Power BI MCP Servers access for an enterprise AI agent programme requires a governance framework that addresses three specific concerns beyond the standard Power BI access controls.

Approved AI client inventory. MCP is an open protocol any MCP-compatible client can connect to the Power BI MCP servers if the connecting user has the required Power BI permissions. Enterprise governance should define which AI clients are approved for MCP server access, and consider whether conditional access policies in Entra ID should restrict the OAuth authentication to approved client applications.

Data classification boundaries. Not all datasets are appropriate candidates for AI agent access via the Remote MCP Server. Datasets containing personally identifiable information, commercially sensitive financial data, or regulated information categories should be assessed against the organisation's data classification policy before MCP access is enabled. Sensitivity labels applied within the Power BI Service provide a starting point for this classification, but the governance decision about which classified data may be processed by external AI models is a policy question that sits above the technical controls.

Audit and activity logging. MCP server operations are mediated by the authenticated user's identity, which means they appear in the Power BI Activity Log as operations performed by that user. Governance teams should ensure that MCP-mediated operations are distinguishable in audit logs from user-initiated operations enabling the monitoring of AI agent activity on Power BI data assets as a distinct category within the organisation's information security audit programme.

Next Steps for AI Agent Integration in Power BI

For enterprise data teams ready to explore Power BI MCP Servers, the starting point is the Modeling MCP Server in a development environment connecting an approved AI client to a non-production semantic model to validate the tool discovery, authentication flow, and DAX generation quality before exposing any production data. This controlled starting point demonstrates value quickly the semantic model documentation and DAX authoring assistance use cases produce visible results within a single development session while containing the governance surface to the development workspace.

The Remote MCP Server introduction should follow the establishment of a data classification review for the target datasets and a confirmed Entra ID conditional access policy for the approved AI clients. For organisations with active Power BI governance programes, these preconditions map directly to existing governance artefacts the dataset certification status, the sensitivity label taxonomy, and the conditional access policies already applied to Power BI Service access.

If your organisation is evaluating how Power BI MCP Servers fit into your AI agent strategy — or needs to design the governance framework that makes responsible MCP access viable for sensitive data environments, speak with a certified Power BI and AI consultant at Numlytics. We work with enterprise analytics and security teams across the US, UK, Australia, and UAE to build AI integration architectures that deliver productivity benefits within the boundaries of enterprise data governance requirements. For the broader context of AI integration in Power BI, see our post on moving from data analytics to AI in Power BI.