MCP Server Guide

Power BI MCP Server: Official Setup and Security Guide

Choose between Microsoft’s hosted remote server and local modeling server, then give your agent the smallest useful set of semantic-model permissions.

The short answer

Microsoft provides official Power BI MCP servers. Both the remote and local/modeling options are currently documented as preview features.

What is the Power BI MCP server?

The Power BI MCP servers let an MCP-compatible agent discover and work with Power BI semantic models through tools defined by Microsoft. The remote server is hosted in Microsoft Fabric, while the local server runs on your machine and is aimed at development and modeling workflows.

This is not a shortcut around Power BI security. The connected identity still needs access to the relevant Fabric workspace and semantic model, and tenant administrators may need to enable the feature. That makes identity design the first setup decision—not the agent prompt.

What you should know before connecting it

Remote MCP server

Microsoft documents a Streamable HTTP endpoint at https://api.fabric.microsoft.com/v1/mcp/powerbi. It uses Microsoft Entra OAuth and runs as a Fabric-hosted service, making it the cleaner choice for managed clients.

Local modeling server

The local option runs over stdio, requires a current Node.js environment, and can authenticate with an Entra user or service principal. It is better suited to desktop development and model-authoring tasks.

Preview means change control

Tool names, tenant switches, and supported operations can evolve during preview. Pin your client configuration, test upgrades, and review Microsoft’s documentation before expanding access.

Setup plan

Choose remote or local — Use remote for centralized hosted access; choose local when the MCP process must run beside your development tools.
Prepare the identity — Create an Entra sign-in or service principal with access only to the required workspace and semantic models.
Confirm tenant settings — Ask the Fabric administrator to enable the relevant MCP capability and verify organizational policies.
Connect the MCP client — Configure Streamable HTTP for the remote endpoint or Microsoft’s documented stdio command for local use.
Test read-only questions first — Validate discovery, metadata, and query behavior before allowing any model-editing operation.

Useful agent workflows

Metric investigation

Let an analyst ask why a KPI moved, inspect the semantic model, and generate a traceable set of follow-up queries.

Model documentation

Summarize tables, measures, relationships, and descriptions into a review document without manually navigating every object.

Development assistance

Use the local modeling server to help draft or inspect model changes, with a human reviewing the resulting diff before publication.

Security checklist

Use a dedicated Entra identity rather than a broad administrator account.
Limit workspace and semantic-model access through Power BI and Fabric roles.
Separate read-only analysis from model-authoring workflows.
Log tool calls and review any write operation before it reaches production.
Revalidate preview behavior after Microsoft or client updates.

Treat every MCP tool as an API capability, not as a harmless chat feature. Start read-only, test in a non-production account, and require human approval for changes.

Frequently asked questions

Does Microsoft have an official Power BI MCP server?

Yes. Microsoft documents both a hosted remote MCP server and a local/modeling MCP server for Power BI, currently in preview.

What is the official remote Power BI MCP endpoint?

Microsoft documents https://api.fabric.microsoft.com/v1/mcp/powerbi as the Streamable HTTP endpoint.

Can an agent bypass Power BI permissions?

No. Access is still governed by Microsoft Entra authentication, tenant settings, and Power BI or Fabric permissions.

Primary documentation

Verify current endpoints, permissions, and preview limitations in the official Microsoft Power BI MCP documentation before production rollout. Vendor capabilities can change faster than third-party guides.

Run an MCP-ready AI agent without server maintenance

OpenClaw Launch handles the agent host, updates, and uptime so you can focus on safe tool design.

Configure your agent