MCP Developer Workshop

Build, deploy, and use an MCP server on OpenShift — from code to production using the full MCP Lifecycle.

What you will do

In this workshop you will take a pre-built stock market MCP server, build a container image, deploy it through the MCP Lifecycle Operator, register it with the MCP Gateway, and test it by speaking the MCP protocol directly with curl. By the end you will have a working MCP server accessible through the governed MCP Gateway — and you will understand every layer of the stack.

Two tracks

Track How it works

Track	How it works
Hands-on track (recommended)	Pre-built server code with rich comments. You read the code, understand the RHOAI 3.5 MCP Lifecycle, then build and deploy using `oc` commands and test with `curl`. ~50 minutes.
Agent-guided track (optional, Stages 1–9)	AgentSherpa, an AI coding agent, walks you through the entire SDLC: intent → design → build → deploy → use. The agent generates code, runs tests, and deploys — you approve each step.

Hands-on track (recommended)

Pre-built server code with rich comments. You read the code, understand the RHOAI 3.5 MCP Lifecycle, then build and deploy using oc commands and test with curl. ~50 minutes.

Agent-guided track (optional, Stages 1–9)

AgentSherpa, an AI coding agent, walks you through the entire SDLC: intent → design → build → deploy → use. The agent generates code, runs tests, and deploys — you approve each step.

Both tracks deploy the same stock market MCP server through the same MCP Lifecycle: MCPServer CR → AI Hub → MCP Gateway.

Your environment

Each participant gets a personal cloud workspace powered by Red Hat OpenShift DevSpaces. Inside that workspace you have:

A terminal with oc, python, pip, and curl pre-installed and pre-authenticated against the cluster
The workshop repository cloned at /projects/mcp-dev-workshop with the server code in server/ and deploy manifests in deploy/
OpenCode Web (optional) — an AI-native IDE with AgentSherpa pre-configured, for the agent-guided track

All you need is a web browser and the credentials your facilitator provides.

Why this matters

AI agents are only as useful as the tools they can reach. Today, connecting an agent to a database, an API, or a business system means writing custom glue code for every integration — different APIs, different auth, different data formats. MCP changes that.

MCP (Model Context Protocol) is an open standard that is rapidly becoming the industry default for agent-to-tool communication. Major AI platforms — Claude, Cursor, VS Code, OpenCode, and others — already support it. Red Hat OpenShift AI 3.5 brings MCP to the enterprise with a full lifecycle: deploy, discover, govern.

In this workshop you go from source code to a production-ready MCP server in under an hour — and you understand every layer of the stack. The skills transfer directly to deploying any MCP server for any domain: internal APIs, databases, monitoring systems, business workflows.

Learning objectives

By the end of the workshop you will be able to:

Explain how MCP separates tool definitions from agent reasoning
Read and understand an MCP server built with the Python SDK v2
Build a container image using OpenShift binary builds
Deploy an MCP server using the MCPServer CR and the MCP Lifecycle Operator
Register a server with the MCP Gateway for federated tool discovery
Speak the MCP protocol (JSON-RPC 2.0) using curl: initialize, tools/list, tools/call

What you will walk away with

By the end of the workshop, every participant will have:

A deployed MCP server on OpenShift — security-hardened, handshake-verified, and visible in the AI Hub catalog
Hands-on experience with the RHOAI 3.5 MCP Lifecycle — from MCPServer CR to MCP Gateway registration
The ability to repeat this for any tool — the same 3 YAMLs and workflow work for any MCP server you build
Protocol-level understanding of MCP — you will have spoken initialize, tools/list, and tools/call by hand, so you know exactly what happens when an AI agent uses a tool
Confidence to evaluate MCP for your own use cases — internal APIs, databases, monitoring, CI/CD, or any system an AI agent should reach

Start here

Read Architecture to understand the MCP Lifecycle layers
Verify your environment in Prerequisites
Hands-on track: Continue to Build & Deploy an MCP Server
Agent-guided track (optional): Go to Stage 1: Welcome & Concepts

The agent-guided track (optional)

If you choose the agent-guided track, AgentSherpa demonstrates a new way of building software. Each stage maps to a phase of the agentic SDLC:

Stage	SDLC Phase	What happens
1. Welcome & Concepts	—	Meet AgentSherpa, learn MCP fundamentals
2. Capture Intent	Intent	Describe what you want to build, agent confirms
3. Skills & Knowledge	Knowledge	Agent loads domain skills, explains each one
4. Design	Design	Pick tools, confirm the server design
5. Plan	Plan	Review the build plan before any code is written
6. Build + Test	Build	Agent generates code, tests each tool with real data
7. Deploy	Deploy + Govern	Build image → MCPServer CR → AI Hub → MCP Gateway
8. Connect & Use	Feedback	Use your tools from AI chat, close the loop
9. Wrap-up	—	Score, certificate, final quiz, cleanup

Stage

SDLC Phase

What happens

1. Welcome & Concepts

—

Meet AgentSherpa, learn MCP fundamentals

2. Capture Intent

Intent

Describe what you want to build, agent confirms

3. Skills & Knowledge

Knowledge

Agent loads domain skills, explains each one

4. Design

Design

Pick tools, confirm the server design

5. Plan

Plan

Review the build plan before any code is written

6. Build + Test

Build

Agent generates code, tests each tool with real data

7. Deploy

Deploy + Govern

Build image → MCPServer CR → AI Hub → MCP Gateway

8. Connect & Use

Feedback

Use your tools from AI chat, close the loop

9. Wrap-up

—

Score, certificate, final quiz, cleanup