TOMPo MCP — Power BI & Fabric Lineage Intelligence

Trace lineage from semantic models → tables → reports → pages → visuals → columns/measures.

An MCP (Model Context Protocol) server that brings Power BI lineage intelligence directly into your AI assistant — GitHub Copilot, Claude Desktop, Cursor, or any MCP-compatible client.

What it does

• Full Lineage: See exactly which columns and measures appear in which visuals, across all reports bound to a semantic model
• Impact Analysis: "What breaks if I rename Employee.StartDate?" — instantly shows every affected visual
• Interactive Visualization: Export a self-contained D3 tree (HTML file) with expand/collapse, zoom, and search
• Zero Infrastructure: Runs locally on your machine using your own Azure identity. No App Service, no Docker, no backend.

Quick Start

1. Install

pip install tompo-mcp

2. Login to Azure

az login

3. Add to VS Code

Option A — mcp.json (recommended)

Create or edit .vscode/mcp.json in your workspace (or ~/.config/Code/User/mcp.json globally):

{
  "servers": {
    "tompo": {
      "command": "python",
      "args": ["-m", "tompo_mcp"],
      "type": "stdio"
    }
  }
}

Option B — settings.json

Add to your VS Code settings.json (Ctrl+Shift+P → "Preferences: Open User Settings (JSON)"):

{
  "mcp": {
    "servers": {
      "tompo": {
        "command": "python",
        "args": ["-m", "tompo_mcp"],
        "type": "stdio"
      }
    }
  }
}

4. Use in Copilot Chat

Open GitHub Copilot Chat and start asking:

> List my Fabric workspaces
> Generate lineage for dataset abc-123 in workspace xyz-456
> What visuals use the Employee.Department column?
> Export the lineage as an interactive HTML file

MCP Tools

How It Works

You type in Copilot Chat
        │
        ▼
Copilot calls TOMPo MCP tools
        │
        ▼
TOMPo runs locally on your machine:
  → Uses your az login identity
  → Calls Fabric REST APIs (getDefinition, Scanner, DAX)
  → Parses model + report definitions
  → Builds lineage tree
  → Returns data to Copilot
        │
        ▼
Copilot shows the lineage tree / impact table
(or opens interactive HTML in your browser)

Authentication

TOMPo uses DefaultAzureCredential which automatically picks up:

1. Azure CLI (az login) — most common for developers
2. VS Code Azure Account — if you're signed into the Azure extension
3. Environment variables — for CI/CD pipelines
4. Managed Identity — for Azure-hosted scenarios

You need access to the Fabric workspaces you want to analyze. No extra app registrations or service principals required.

Metadata Extraction (3-tier fallback)

1. Fabric getDefinition API — returns full model.bim / TMDL / PBIR definitions
2. Admin Scanner API — fallback if getDefinition fails (requires admin permissions)
3. DAX executeQueries — last resort using INFO.TABLES(), INFO.COLUMNS(), etc.

If a sensitivity label blocks access, TOMPo temporarily downgrades to "General", extracts metadata, then restores the original label.

Interactive Visualization

The export_lineage_html tool generates a single HTML file with:

• D3 horizontal tree with expand/collapse nodes
• Color-coded by type (model, table, report, page, visual, column, measure)
• Impact Analysis tab with searchable data grid
• Semantic Model tab with tables, columns, measures, relationships
• Zoom, pan, fullscreen — all interactive
• Works offline — all JavaScript and CSS inlined, no server needed

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "tompo": {
      "command": "python",
      "args": ["-m", "tompo_mcp"]
    }
  }
}

Cursor / Windsurf

{
  "mcpServers": {
    "tompo": {
      "command": "python",
      "args": ["-m", "tompo_mcp"]
    }
  }
}

Development

git clone https://github.com/microsoft/tompo-mcp.git
cd tompo-mcp
pip install -e ".[dev]"
python -m pytest tests/

Requirements

• Python 3.10+
• Azure CLI (az login) or any Azure credential
• Access to Fabric/Power BI workspaces

License

MIT