Microsoft Showcases Real-World Use of Model Context Protocol (MCP) In Vertical Industry App

As vendors, partners, and customers increasingly adopt Model Context Protocol (MCP), AI agents gain more efficient access to a wide range of data sources.

Against that backdrop, Microsoft has published an MCP-powered sample application that delivers at least two valuable insights: a clear-cut, real-world example of how an enterprise can put MCP into practice, and how MCP can streamline a cumbersome process that operates across a range of systems.

The Microsoft Developer Community Blog details the sample application, including the use of MCP, an agentic orchestration system, and Azure software to support applications on a serverless basis.

Business Challenge, Tech Platforms

The sample application, called AI Travel Agents, shows how developers can orchestrate multiple AI agents (including those written in a diverse set of languages) to perform a complex task, in this case creating trip scenarios for customers of a travel planning firm.

The application is built to overcome the complexity inherent in analyzing diverse customer needs, recommending destinations, and crafting itineraries, all while integrating real-time data such as trending locations or logistics that impact a travel plan. This process suffers from latency in traditional systems; AI agents aim to coordinate information and plans in real time.

AI Travel Agents is built on three main components:

LlamaIndex.TS, an open-source framework for building agentic generative AI applications connected to enteprrise data sources, which in this case orchestrates AI agents for efficient task handling
MCP, which provides travel-specific data from various systems to travel agents
Microsoft Azure Container Apps provides scalable, serverless deployment and dynamic workload management

LlamaIndex.TS orchestrates interactions and functions including:

Task Delegation: the Triage Agent analyzes queries and routes them to specialized agents, such as the Itinerary Planning Agent, for efficient workflow
Agent Coordination: LlamaIndex.TS maintains context across interactions, enabling coherent responses to complex queries; one good example is a set of plans for a trip that includes multiple cities
Large Language Model (LLM) Integration: It connects to Azure OpenAI, GitHub Models and other LLMs with advanced AI functionality

LlamaIndex.TS’s modular design means new agents can be added in a simple manner to extend the software’s functionality.

MCP provides real-time, travel-specific data, such as trending destinations or seasonal events through Bing search’s Web Search Agent, as well as connecting agents to external tools including Python-based itinerary planning for trip schedules written in Java.

One example of MCP in action: when the Destination Recommendation Agent needs current travel trends, MCP delivers them via the Web Search Agent. While LlamaIndex.TS’s main function is orchestration, MCP is enriching agent capabilities through system connections that can be added as/when needed. MCP provides the data needed to tailor recommendations to a travel agency’s client and their needs.

Azure Container Apps powers the sample application with a serverless platform to deploy microservices and effectively manage varying workloads. Azure Container Apps perform dynamic scaling for managing demand surges, support for microservices including .NET, Python, Java, and Node.js; and observability functions including, tracking, metrics, and logging.

AI Travel Agents leverages Azure Container Apps’ real-time chat for efficient responses for agents, modularity of MCP to support future additions, and a UI that displays agent reasoning for debugging and transparency into the insights that are delivered.

Closing Thoughts

This sample application is a valuable addition to the understanding, development of use cases, and planning around MCP. As such, it’s a solid resource for customers to advance their analysis of MCP functionality and the opportunies presented by this specification.

Developers, customers, or interested stakeholders can try out a live demo for free — at GitHub — to gain real-world insights and experiences.

Additional MCP, AI Agent Interoperability Analysis: