Fabric Sales Agent Accelerator

📖 Full documentation — architecture, demo script, setup guide, costs

What it is

An AI sales agent accelerator for Wide World Importers that shows how to pair a shared Microsoft Fabric Data Agent backend with two delivery surfaces: a fast developer prototype in GitHub Copilot CLI and a production path into M365 Copilot + Teams.

The Two-Surface Approach

• Surface 1: GitHub Copilot CLI — prototype quickly with MCP servers for Fabric data, M365 activity context, and inline report generation.
• Surface 2: M365 Copilot + Teams — graduate the same business flow into an Azure AI Foundry agent published through an Agent Application.
• WorkIQ note: production uses WorkIQ for M365 activity context; the demo tenant uses a mock WorkIQ tool with sample activity data until tenant provisioning is available.

Architecture diagram

┌─ CLI Surface (Prototype) ─────────────────────┐
│ Copilot CLI → MCP Servers                      │
│   → wwi-sales-data (Fabric Data Agent)         │
│   → workiq (M365 activity data)                │
│   → quota-forecast skill (inline report)       │
└────────────────────────────────────────────────┘

┌─ M365 Surface (Production) ───────────────────┐
│ M365 Copilot / Teams → Foundry Agent           │
│   → Fabric IQ (NL→SQL platform tool)           │
│   → WorkIQ (M365 activity platform tool, OBO)  │
│   → Report Generator (DOCX + OneDrive link)    │
└────────────────────────────────────────────────┘

Same Data Agent backend. Same business logic. Different distribution.

Quick Start (CLI)

git clone https://github.com/ericchansen/agent-demo.git && cd agent-demo
uv sync --extra dev              # or: python -m venv .venv && pip install -e ".[dev]"

Edit .github/mcp.json — replace <your-workspace-id> with your Fabric workspace GUID. Then:

copilot
# > What were Tailspin Toys' total sales last quarter?

See the full setup guide for all options (uv, pip+venv, copilot mcp add, Foundry surface).

Tech Stack

• Microsoft Fabric — shared data agent and OneLake analytics
• Azure AI Foundry — production agent for M365 Copilot and Teams
• Model Context Protocol (MCP) — tool surface for the CLI prototype
• Python 3.11+ — agent, tool, and report-generation implementation

Cost

Plan for roughly ~$270/month active for the demo footprint. Pause Fabric capacity when idle to drop to roughly ~$15/month in residual costs. See docs/costs.md.

Repository Structure

Path	Purpose
src/cli/	GitHub Copilot CLI prototype surface: MCP config and skills
src/orchestrator/	Azure AI Foundry agent for the M365 Copilot + Teams surface
src/agents/	Local MCP servers, demo mocks, and report generation helpers
fabric/	Fabric Data Agent instructions and example queries
infra/	Bicep infrastructure for Fabric and Azure resources
demo/	Wide World Importers sample assets and demo content
docs/	Setup, architecture, cost, and two-surface guidance
docs/surfaces/	Reference-only alternatives such as Copilot Studio and M365 Direct Publish
tests/	Unit and integration coverage

License

MIT