Skip to content

Proxuma/powerbi-copilot

BranchesTags

Repository files navigation

Proxuma Power BI Copilot

Generate AI-powered reports and dashboard builders directly from your Power BI data using GitHub Copilot, Claude, or ChatGPT.

This package connects your AI assistant to Power BI via MCP (Model Context Protocol). You ask a business question, the AI queries your data model, and outputs a complete HTML report or dashboard builder with real numbers.

What's included

Component Description
MCP Server Python server that connects AI tools to Power BI and Microsoft Fabric APIs
Report Prompt Generates standalone HTML reports with KPIs, data tables, analysis, and actionable findings
Dashboard Prompt Generates interactive dashboard mockups with step-by-step Power BI build instructions
PII Anonymization All data passes through Microsoft Presidio before reaching the AI — names, emails, and other PII are automatically masked
Setup Wizard Auto-discovers workspaces and datasets — no manual GUID hunting

Data Anonymization

All data that passes through the MCP server is automatically anonymized before it reaches the AI. Real names, emails, and other PII are replaced with consistent aliases (Client_A, Resource_1, etc.) that the AI uses in its output.

How it works

  1. On first query, the server loads all unique values from your configured sensitive columns (company names, resource names, contacts)
  2. Every response passes through two anonymization layers:
    • Deterministic lookup: known entities get consistent aliases
    • Presidio NLP: catches unexpected PII in free-text fields
  3. After report generation, restore real names locally:
# CLI
python -m server report.html -o report-real.html

# Or drag-and-drop mapping.json onto the report page

Configuration

Edit ~/.powerbi-mcp/config.json:

{
  "anonymization": {
    "enabled": true,
    "sensitive_columns": {
      "client": ["'YourTable'[Company Name]"],
      "resource": ["'YourTable'[Full Name]"],
      "contact": ["'YourTable'[Contact Name]"]
    },
    "presidio_enabled": true
  }
}

The setup wizard (python wizard.py) auto-detects likely sensitive columns and lets you confirm.

Audit Trail

Every session stores its mapping at ~/.powerbi-mcp/sessions/<id>/mapping.json. This file never leaves your machine. Use it to:

  • Verify exactly what was anonymized
  • Restore real names in reports
  • Provide compliance documentation

Requirements

  • Python 3.10+
  • Power BI Pro or Premium Per User license (for API access)
  • VS Code with GitHub Copilot (Agent mode) or another MCP-compatible AI tool

No Azure app registration needed. The server uses Azure's public client ID for authentication.

Quick Start

One-liner install (macOS / Linux)

curl -sL proxuma.io/install | bash

One-liner install (Windows PowerShell)

irm proxuma.io/install.ps1 | iex

Manual install

# 1. Clone the repo
git clone https://github.com/Proxuma/powerbi-copilot.git
cd powerbi-copilot

# 2. Run the installer — it installs deps and launches the setup wizard
./setup.sh          # macOS/Linux
.\setup.ps1         # Windows PowerShell

# 3. The wizard opens your browser for Microsoft sign-in,
#    then lets you pick your workspace and dataset.
#    Config is written automatically.

# 4. Open in VS Code, start Copilot Chat (Agent mode), type:
#powerbireport what is my monthly revenue trend?

That's it. No config files to edit, no GUIDs to hunt for.

Enterprise Deployment

For IT admins deploying to multiple machines via MDM (Intune, JAMF, GPO):

# Pre-configured (skip interactive wizard)
./setup.sh --workspace-id "GUID" --dataset-id "GUID" --silent

# Config from IT-hosted endpoint
./setup.sh --config-url https://it.yourcompany.com/powerbi-mcp-config

# Headless / SSH environments
./setup.sh --device-code --workspace-id "GUID" --dataset-id "GUID"

Windows (Intune):

.\setup.ps1 -WorkspaceId "GUID" -DatasetId "GUID" -Silent

Environment variables (override config.json — set via GPO/Intune):

  • POWERBI_MCP_WORKSPACE_ID — default workspace
  • POWERBI_MCP_DATASET_ID — default dataset
  • POWERBI_MCP_CONFIG — full JSON config blob

Available MCP Tools

Once the server is running, your AI assistant has access to these tools:

Tool Description
list_workspaces List all Power BI workspaces you have access to
list_datasets List all datasets in a workspace
execute_dax Run a DAX query and get results
search_schema Search for specific measures, columns, or tables (recommended)
list_measures List all measure names in a dataset
list_fabric_items List all items in a Fabric workspace
get_schema Get the full schema (caution: can be >10MB)

Prompt Files

#powerbireport

Generates a self-contained HTML report from your Power BI data. The output includes:

  • Summary KPIs
  • Data tables with real query results
  • Collapsible DAX queries (copy-paste into Power BI Desktop)
  • Analysis narrative
  • Actionable findings
  • FAQ section

#powerbidashboard

Generates an interactive dashboard mockup with build instructions. The output includes:

  • Visual preview with page tabs and sample data
  • 15-25 DAX measures with copy buttons
  • Step-by-step build instructions (written for beginners)
  • Conditional formatting color rules

Dashboard Renderer

The Dashboard Renderer is a self-contained HTML page where users paste JSON output from the AI agent to get a full interactive dashboard. The AI agent outputs two JSON objects (DASH_CONFIG + DATA), and the renderer builds the dashboard client-side.

How it works

1. User asks: "Generate a dashboard for SLA performance"
2. AI agent queries Power BI via MCP tools
3. Agent outputs DASH_CONFIG + DATA as JSON (fits within token limits)
4. User opens the Dashboard Renderer
5. User pastes JSON → live dashboard with KPIs, sparklines, charts, and tables
6. Optional: download as standalone HTML file

Pre-built Template Library

8 ready-to-use DASH_CONFIG templates in templates/configs/:

Template File KPIs
SLA Performance sla-performance.json FR SLA %, Resolution SLA %, Overdue
Ticket Volume ticket-volume.json Created, Resolved, Open, Backlog
Revenue & Billing revenue-billing.json Revenue, Hours, Avg Rate, Margin
Resource Utilization resource-utilization.json Billable %, Total Hours, Non-Billable
CSAT & Satisfaction csat-satisfaction.json Avg CSAT, Response Rate, Surveys
Patch Compliance patch-compliance.json Patched %, Critical, Overdue
Endpoint Health endpoint-health.json Online %, Alerts, Tickets
Project Performance project-performance.json Budget %, Actual Hours, Milestones

Hosting Options

Three tiers for hosting the Dashboard Renderer, from simplest to most integrated:

Option A: Proxuma-Hosted (SaaS) — Free

The easiest path. Proxuma hosts the renderer, MSPs just paste JSON.

  • URL: proxuma.io/powerbi/dashboards/renderer/
  • Template gallery: proxuma.io/powerbi/dashboards/ — pick a template, paste your data
  • Cost: Free (included with Proxuma)
  • Setup: None — just open the URL
  • Data privacy: JSON is processed client-side only; no data sent to Proxuma servers

Option B: Azure Static Web App (Customer Tenant) — Free

MSPs deploy their own renderer to their Azure tenant. Data never leaves their environment.

cd azure
./deploy-static.sh --name mycompany-dashboards
  • URL: https://mycompany-dashboards.azurestaticapps.net/dashboard-renderer.html
  • Cost: Free (Azure Static Web Apps Free tier — 100 GB bandwidth/month)
  • Setup: 1 command, ~2 minutes
  • Benefits: Custom domain, Azure AD auth possible, data stays in tenant
  • Files: azure/deploy-static.sh, azure/staticwebapp.config.json

Option C: Fabric / Power BI Embedded (Advanced)

For MSPs already on Microsoft Fabric who want native integration.

Approach Description Complexity
SharePoint page Embed the Azure Static Web App in an iframe on a SharePoint page Low
Power BI paginated report Embed dashboard HTML as a paginated report Medium
Power BI custom visual Custom visual that renders DASH_CONFIG/DATA JSON High

This is the most integrated option but requires Fabric capacity (F2+) or Power BI Premium.

Pricing Comparison

Proxuma-Hosted Azure Static Web App Fabric Embedded
Monthly cost Free Free (100 GB/mo) Fabric F2+ required
Setup time 0 min 2 min 30+ min
Custom domain No Yes N/A
Azure AD auth No Optional Built-in
Data in tenant Client-side only Yes Yes

Project Structure

powerbi-copilot/
├── server/
│   ├── server.py            # MCP server
│   ├── auth.py              # Shared authentication module
│   ├── wizard.py            # Setup wizard (auto-discovery)
│   ├── requirements.txt     # Python dependencies
│   └── config.example.json  # Example configuration
├── prompts/
│   ├── powerbireport.prompt.md    # Report generator prompt
│   └── powerbidashboard.prompt.md # Dashboard builder prompt
├── templates/
│   ├── dashboard-renderer.html    # Dashboard Renderer (v2)
│   ├── dashboard-shell.html       # Dashboard builder template
│   ├── report-shell.html          # Report template
│   ├── configs/                   # Pre-built DASH_CONFIG templates
│   │   ├── sla-performance.json
│   │   ├── ticket-volume.json
│   │   ├── revenue-billing.json
│   │   ├── resource-utilization.json
│   │   ├── csat-satisfaction.json
│   │   ├── patch-compliance.json
│   │   ├── endpoint-health.json
│   │   └── project-performance.json
│   ├── mcp.json             # VS Code MCP config template
│   └── settings.json        # VS Code settings template
├── azure/
│   ├── deploy.sh            # MCP server Azure Functions deployment
│   ├── deploy-static.sh     # Dashboard Renderer Static Web App deployment
│   ├── staticwebapp.config.json  # SWA route configuration
│   ├── host.json            # Azure Functions host config
│   └── requirements.txt     # Azure Functions Python deps
├── experiments/
│   └── RESULTS.md           # Copilot Studio experiment results
├── docs/
│   ├── first-run.md         # First-time setup & auth
│   └── troubleshooting.md   # Common issues & fixes
├── setup.sh                 # macOS/Linux installer
├── setup.ps1                # Windows PowerShell installer
├── install.sh               # One-liner web installer (macOS/Linux)
├── install.ps1              # Windows full installer
└── README.md

How Authentication Works

The server uses Azure AD's public client flow — the same one Power BI Desktop uses. No app registration required.

  1. The setup wizard (or first MCP tool call) opens a browser for Microsoft login
  2. You sign in with your Power BI account
  3. Tokens are cached in ~/.powerbi-mcp/
  4. Subsequent runs: no login needed (tokens refresh automatically)

The cached tokens are only stored on your machine. The MCP server never sends credentials to any third party.

Compatibility

AI Tool Status
Microsoft Copilot Studio Supported (tested Mar 2026, JSON dashboard output)
GitHub Copilot (VS Code, Agent mode) Supported
Claude Code (CLI) Supported
Claude Desktop Supported (add to claude_desktop_config.json)
Cursor Supported (add to MCP settings)
ChatGPT (via MCP plugin) Experimental

License

Proprietary. Access by invitation only.

About

Power BI MCP server + AI prompt files for Copilot/Claude/ChatGPT — generate reports and dashboards from your Power BI data

Resources

Readme

License

View license
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages