Power BI Desktop MCP Server

⚠️ Alpha version – APIs and behaviours may change without notice.

Power BI Desktop MCP Server lets Claude (or any MCP client) talk to a running Power BI Desktop instance using natural language. It bridges the Model Context Protocol to the Tabular Object Model so you can inspect, optimize, and author DAX without leaving chat.

Installation

Step-by-step setup (Power BI prerequisites, .NET libraries, environment variables) lives in INSTALLATION.md.

For local development:

pip install -e .
# or install from the current directory:
pip install .

Quick Example

List the tables in your current Power BI model:

python examples/list_model_info.py --include-columns

Export measures to the knowledge base:

python examples/export_measures_to_knowledge_base.py --project "Executive Dashboard"

Run the optimizer on an expression:

python examples/optimize_dax.py --expr "CALCULATE(SUM(Sales[Amount]), FILTER(ALL(Date), Date[Year] = 2024))"

Configuration

• Copy .env.example to .env and set TOM_DLL_PATH if the Microsoft Analysis Services DLLs are not auto-detected. Optional variables include LOG_LEVEL, MCP_KNOWLEDGE_BASE, and POWERBI_WORKSPACE_NAME.

• Add the CLI entry point to Claude Desktop (or any MCP host):

  {
    "mcpServers": {
      "powerbi": {
        "command": "powerbi-mcp-server",
        "args": []
      }
    }
  }
  

• The CLI is a thin wrapper around python -m powerbi_mcp.server, so either invocation works for deployments and debugging.

Features

• DAX code generation and measure creation
• Measure and table metadata inspection
• DAX optimizer with rule-based suggestions
• TMDL import/export helpers
• Knowledge base bridge for storing curated measures
• Example scripts for quick automation (examples/ directory)

Known Issues

• Power BI visuals live outside the XMLA endpoint, so visual creation/editing is not supported.
• The server connects to the most recently opened Power BI Desktop workspace; multiple instances are not yet multiplexed.
• Write operations require the Power BI model to allow external tools and, in some scenarios, elevated permissions.
• pythonnet relies on local Microsoft Analysis Services DLLs; misconfigured TOM_DLL_PATH is the most common setup failure.

Contributing

Pull requests are welcome. Please:

1. Open an issue describing the change if it alters behaviour or APIs.
2. Keep additions formatted with ruff/black style conventions (if configured) and include type hints.
3. Add or update example scripts/tests where it helps illustrate new functionality.
4. Run the example scripts or manual smoke tests before submitting.

Issues and feature requests are tracked at the repository’s GitHub Issues page.