Skip to main content Link Menu Expand (external link) Document Search Copy Copied
L300 Extending M365

Overview

Declarative agents are specialized, context-aware Copilots designed to streamline work by addressing specific workflows, context and business needs. When you create a declarative agent, you’re building on and extending Microsoft’s AI stack.

A declarative agent application package typically consists of the following files

  • An app manifest in JSON (the standard teams app manifest file called manifest.json)
  • A declarative agents manifest in JSON which will consist of the agent’s name, instructions, capabilities, conversation starters and actions if needed.
  • An optional plugin manifest in JSON to configure your action as an API plugin if they have authentication, required fields, adaptive card responses etc. This file is only needed if you have an action in your agent.

Exercises in this lab

You’ll explore the fundamentals of declarative agents and how you can integrate them with MCP servers. You’ll then build a declarative agent that uses multi‑agent orchestration to coordinate multiple agents and deliver a unified response to a user’s prompt.

In this lab:

  • Introduction
  • Exercise 01: Connect declarative agent to MCP Server
  • Exercise 02: Multi-agent claims orchestration

Introduction

What you’ll build

Zava Insurance is a fictional mid-sized insurance company that serves over 150,000 homes across the Pacific Northwest. Like many companies, they faced a challenge: their manual claims process was slow and inefficient. When severe storms hit in October 2025, they received 2,000 claims in just 48 hours-and it took weeks to process them all.

Problem

Manual processing meant:

  • 3-week delays for customers waiting on claims.
  • Coordination issues when handling multiple urgent cases.
  • Overwhelmed staff, unable to focus on complex cases that need human expertise.

Solution

Zava’s development team decided to build an AI-powered claims system that would:

  1. Automate routine tasks, so adjusters can focus on what matters most.
  2. Provide instant access to data through natural language conversations.
  3. Work seamlessly with tools employees already use (like Microsoft 365).

To make this happen, they built two key components:

🔧 MCP Server (Model Context Protocol Server)

  • Think of this as a “translator” that lets AI agents securely access claims data.
  • Built on Azure infrastructure for security and reliability.
  • Provides real-time information about damage assessments, contractors, and inspection schedules.

💬 Declarative agents with Microsoft 365 Copilot

  • Allows adjusters to use simple, natural language like “Show me all urgent storm damage claims”.
  • No need to learn complex systems or navigate multiple databases.
  • Works right inside Microsoft 365 tools they use every day.

Lab 2 objectives

By the end of this lab, you’ll:

  • Understand how MCP servers bridge the gap between AI agents and backend systems.
  • Build and run Zava’s MCP server with real insurance claims data.
  • Create declarative agents using Microsoft 365 Agents Toolkit that:
    • Connect to your MCP Server.
    • Access knowledge and data from SharePoint and other sources.
  • Build an orchestrator agent that coordinates multiple specialized agents.
  • Test your agent using natural language queries with real claims data.