使用说明

Power BI MCP Server

A remote Model Context Protocol (MCP) server that exposes Power BI operations as tools for AI assistants. Built with the official Microsoft MCP C# SDK and served over Streamable HTTP.

Portfolio summary

What it demonstrates: A typed, remote MCP integration over enterprise analytics rather than a local toy server.
Tools: Workspace and dataset discovery, schema inspection, and DAX query execution.
Identity: DefaultAzureCredential supports local development, service principals, workload identity, and managed identity.
Engineering: .NET 8, Streamable HTTP, structured error handling, tests, and client configuration for VS Code.

Tools

Tool	Description	Parameters
`list_workspaces`	List all Power BI workspaces accessible to the authenticated user	—
`list_datasets`	List datasets in a workspace	`workspaceId`
`list_tables`	List tables and columns in a dataset (via DAX `COLUMNSTATISTICS()`)	`workspaceId`, `datasetId`
`execute_dax`	Execute a DAX query via REST API	`workspaceId`, `datasetId`, `daxQuery`

Prerequisites

.NET 8 SDK
ASP.NET Core 8 Runtime (aspnetcore-runtime-8.0)
Azure CLI (az login) or a service principal for authentication
A Power BI Pro/PPU/Premium workspace

Quick Start

# 1. Clone and enter the repo
git clone <repo-url> && cd pbi-mcp

# 2. Log in to Azure (use your Power BI tenant)
az login --tenant <your-tenant-id>

# 3. Update appsettings.json with your tenant ID
#    (already configured if you cloned this repo)

# 4. Build and run
dotnet run

The server starts at http://localhost:3001/mcp.

Configuration

appsettings.json

{
  "PowerBi": {
    "TenantId": "your-tenant-id-here"
  }
}

The TenantId is passed to DefaultAzureCredential to ensure tokens are acquired for the correct tenant.

Authentication

The server uses DefaultAzureCredential, which tries these sources in order:

Priority	Source	Use case
1	`AZURE_TENANT_ID` / `AZURE_CLIENT_ID` / `AZURE_CLIENT_SECRET` env vars	Service principal / CI
2	Workload Identity	Kubernetes
3	Managed Identity	Azure VMs, App Service, ACA
4	Azure CLI (`az login`)	Local development

Local development (az login)

az login --tenant <your-tenant-id>
dotnet run

Service principal (environment variables)

export AZURE_TENANT_ID="your-tenant-id"
export AZURE_CLIENT_ID="your-sp-app-id"
export AZURE_CLIENT_SECRET="your-sp-secret"
dotnet run

No code changes needed — DefaultAzureCredential picks up the env vars automatically.

Connecting an MCP Client

VS Code (Copilot)

The repo includes .vscode/mcp.json:

{
  "servers": {
    "pbi-mcp": {
      "type": "http",
      "url": "http://localhost:3001/mcp"
    }
  }
}

Start the server with dotnet run, then Copilot will discover the tools automatically.

MCP Inspector

npx @modelcontextprotocol/inspector

Connect to http://localhost:3001/mcp using the Streamable HTTP transport.

Project Structure

PbiMcpServer.csproj          # ASP.NET Core Web project
Program.cs                   # Entry point — DI, MCP server, HTTP transport
PowerBiService.cs            # Power BI REST API + XMLA service layer
PowerBiTools.cs              # MCP tool definitions (4 tools)
appsettings.json             # Tenant ID + logging config
.vscode/mcp.json             # VS Code MCP client config
tests/
  PbiMcpServer.Tests/
    PowerBiServiceUnitTests.cs        # 11 offline tests (mocked HTTP)
    PowerBiServiceIntegrationTests.cs # 5 live API tests
    XunitLoggerProvider.cs            # xUnit ↔ ILogger bridge

Testing

# Unit tests (offline, fast — no credentials needed)
dotnet test tests/PbiMcpServer.Tests/PbiMcpServer.Tests.csproj \
  --filter "Category!=Integration"

# Integration tests (hits live Power BI API — requires az login)
dotnet test tests/PbiMcpServer.Tests/PbiMcpServer.Tests.csproj \
  --filter "Category=Integration"

# All tests
dotnet test tests/PbiMcpServer.Tests/PbiMcpServer.Tests.csproj

# With detailed HTTP request/response logging
dotnet test tests/PbiMcpServer.Tests/PbiMcpServer.Tests.csproj \
  --filter "Category=Integration" \
  --logger "console;verbosity=detailed"

Architecture Notes

DAX execution

All DAX queries use the Power BI REST API (POST /datasets/{id}/executeQueries). This works with any Pro/PPU/Premium workspace — no XMLA endpoint required.

The XMLA path (via ADOMD.NET) is included in the codebase but commented out. It requires Premium/PPU capacity with XMLA read enabled by a tenant admin.

Table discovery

The list_tables tool uses EVALUATE COLUMNSTATISTICS() via the REST DAX endpoint rather than the REST /tables endpoint, because /tables only works for push datasets. The DAX approach works for all dataset types (import, DirectQuery, composite).

.NET runtimes

This project requires both:

Microsoft.NETCore.App — base .NET runtime (console, I/O, networking)
Microsoft.AspNetCore.App — web runtime (Kestrel, routing, middleware)

Install with: sudo apt install aspnetcore-runtime-8.0 (includes both).

Tech Stack

.NET 8 / ASP.NET Core
ModelContextProtocol.AspNetCore — Official MCP C# SDK with Streamable HTTP transport
Azure.Identity — DefaultAzureCredential for token acquisition
Microsoft.AnalysisServices.AdomdClient — XMLA/DAX connectivity
xUnit + Moq — Testing

License

This project is licensed under the MIT License.