workshop-27-08-26

Example MCP server for the AWS AI FDE for Commerce bootcamp workshop on 27/08/2026

What is MCP?

The MCP (Model Context Protocol) is an open-source communication protocol between systems that allows the exchange of data/context information between different applications (mostly AI-based).

Using this protocol, an AI system (such as an agent application) can consume/provide contextual information from/to another system, allowing models to operate with more certainty and relevance.

Using the USB-C analogy:

Think of MCP as a USB-C port for AI applications. Just as USB-C provides a standardized way to connect electronic devices, MCP provides a standard way to connect AI applications to other systems.

Related MCP server: Basic MCP Server

SDK Options

The protocol supports different SDK (Software Development Kit) options, which can be used to implement MCP in different languages/frameworks.

Find the full list here: Available SDKs

Transport Options

Focusing on the workshop context, let's explore the most common/popular transport options for Node.js:

stdio

Run the MCP server as a child process, sending and receiving messages via stdin and stdout.

Serve over stdio

HTTP

Provide a server that receives and sends messages via HTTP requests to multiple clients simultaneously.

Serve over HTTP

Frameworks

Use Node.js technologies/frameworks to make implementing the MCP protocol easier, such as:

• Express

• Fastify

• Hono

Workshop Scenario

Client Nordesul

1. Consider an e-commerce project for a major client, with a team of 20 developers assigned to migrate their legacy virtual store (V0) to a more modern framework (V1). The goal is for the V1 virtual store to be identical to V0, but using the performance improvements of V1.

2. Since multiple developers will work on the same codebase, it is essential that deliverables have the best possible standardization and quality, following our company's standards and identity.

3. The client has a database with documentation for all their APIs, available at a shared endpoint for the development team to consult.

This Repository

This repository contains an MCP server template that can be used to:

• Understand the implementation of stdio and HTTP transports with @modelcontextprotocol/server;

• Simulate communication between AI systems and any external service;

• Serve as inspiration for implementing a real MCP server;

• Be the gateway to using AI Agents with custom tools.

Repository Structure

/servers

Example MCP servers, using different transports and frameworks.

• /servers/stdio

Serves MCP messages via stdin and stdout, as a child process.

• /servers/http

Serves MCP messages via HTTP requests, using only Node.js's native HTTP module.

• /servers/express

Serves MCP messages via HTTP requests, using the Express framework.

/tools

Examples of custom tools that can be used by AI Agents to interact with other systems.

• /tools/index.js

Centralizes and exports all tools in the repository, so MCP servers can register them in a unified way.

• /tools/nordesul-deploy

Returns essential care and best practices for a safe deploy at the Nordesul client.

• /tools/nordesul-delivery

Simulates implementation patterns and contracts for the project.

• /tools/nordesul-reference

Simulates querying an API documentation at the Nordesul client.

• /tools/nordesul-status

Simulates querying the status of an application at the Nordesul client.

/utils

Centralizes the project's utility/shared functions.

/.opencode

Opencode configuration directory, contains a custom mentor agent to help workshop participants understand the repository and its features.

Running the Project

Prerequisites

• Git;

• Clone the workshop repository;

• Opencode (or any other AI agent that supports MCP) installed and configured;

• Node.js (>= LTS preferred)

• Package Manager (npm or yarn)

Installing Dependencies

At the project root, run the command:

npm install

# ou:

yarn install

Running the Servers

MCP Server via stdio:

npm run start:stdio

# ou:

yarn start:stdio

MCP Server via http:

npm run start:http

# ou:

yarn start:http

MCP Server via Express integration:

npm run start:express

# ou:

yarn start:express

Running the Mentor Agent (Example with Opencode)

With Opencode installed, open a terminal of your choice in the repository directory and run the command:

opencode

The first run will download all Opencode dependencies, so wait for the process to complete. The Mentor agent will already be selected by default.

Integrating MCP servers with AI Agents

• Opencode: MCP servers;

• Claude: Connect Claude Code to tools via MCP;

• Codex: Model Context Protocol;

• Cursor: Model Context Protocol (MCP);

• Antigravity: Model Context Protocol (MCP).

Example Prompts

Some examples for interacting with the repository's tools, using your agent:

nordesul-delivery

Quais são os padrões de implementação para a área de checkout do Nordesul?

nordesul-deploy

Antes de eu publicar essa mudança no Nordesul, quais cuidados devo tomar?

nordesul-reference

Me mostra a documentação da API de pedidos (orders) do Nordesul.

nordesul-status

Qual o status atual da aplicação checkout-service no Nordesul?