Copilot Insights

Governance and telemetry reporting for Microsoft Copilot Studio agents across an M365 tenant. Pulls data from a dozen Azure and Microsoft 365 APIs, stores it in a local SQLite database, and outputs a multi-sheet Excel workbook — plus an MCP server that lets AI agents (Copilot Studio, M365 Copilot, AI Foundry) query the telemetry in plain English.

What it does

• Collects usage, connector health, DLP violations, agent inventory, and M365 Copilot adoption from across your tenant

• Stores everything in a local SQLite database (optionally synced to Azure Blob Storage)

• Exports a timestamped Excel workbook with 20+ analytical sheets

• Exposes an MCP server (stdio locally, HTTP on Azure Container Apps) so AI agents can query the data conversationally

Related MCP server: Fabric-Analytics-MCP

Data sources

Excel output sheets

Summary, Agents, Invocations, Connectors, Environments, Publishers, DLP Policies, AI Usage, Azure Health, Cross-Reference, KPI History, M365 Copilot, M365 Copilot Trend, M365 App Users, O365 Users, M365 Packages, Teams Usage, Viva, Viva Sessions, Viva Topics, Viva WAU, Viva Autonomous

Prerequisites

• Python 3.12+

• Azure CLI (az login) — or a service principal with the permissions granted by provision/step2_identity.sh

• An M365 tenant with Copilot Studio agents and Application Insights connected

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

Configuration

Copy the example and fill in your values:

cp config/.env.example .env

Key variables:

For multi-tenant use, keep a separate .env file per tenant and pass it with --env:

python -m src.main --env .env.mwc all

Per-datasource auth overrides (e.g. LOG_ANALYTICS_CLIENT_ID) are documented in config/.env.example.

Usage

# Fetch all sources and export workbook in one step
python -m src.main all

# Fetch only (no export)
python -m src.main sync

# Export last sync to Excel (no new fetch)
python -m src.main export

# Import Copilot Studio analytics CSV exports from Viva
python -m src.main import-viva <path/to/csv/folder>

The workbook is written to OUTPUT_PATH (default: agent_telemetry_<timestamp>.xlsx).

MCP server

The MCP server lets AI agents (Copilot Studio, M365 Copilot, Azure AI Foundry) query telemetry in natural language.

Local (Claude Code / stdio)

Add to your .mcp.json:

{
  "mcpServers": {
    "agent-telemetry": {
      "command": "python",
      "args": ["-m", "src.mcp_server.server"]
    }
  }
}

Deployed (Azure Container Apps / HTTP)

bash deploy.sh deploy-config/<tenant>.env

The deploy script provisions an ACR, builds and pushes the Docker image, creates an Azure Container App, and wires up Entra ID authentication. Microsoft agent platforms send Bearer tokens automatically.

Provisioning a new tenant

Run the three provision scripts in order (from Azure Cloud Shell or with appropriate admin roles):

# 1. Create Log Analytics workspace + Application Insights
bash provision/step1_insights.sh

# 2. Grant the sync service principal all required read permissions + admin consent
bash provision/step2_identity.sh

# 3. Register the MCP server app in Entra ID
bash provision/step3_mcp.sh

See the header comments in each script for required admin roles and outputs.

Project structure

src/
  fetchers/       # One module per data source
  writers/        # One module per Excel sheet
  mcp_server/     # MCP server (stdio + HTTP)
  agent/          # Bot Framework conversational agent
  store/          # SQLite + Azure Blob Storage
config/           # Settings and datasource config (gitignored — use .env)
provision/        # One-time Azure setup scripts
deploy.sh         # Build + deploy to Azure Container Apps
Dockerfile        # Container image for the MCP server