Skip to content
MCP
Beginner-friendly · Production-grade

Build a production-ready MCP server for Microsoft Copilot Studio

Learn to design, build, secure, test, deploy, and operate a Model Context Protocol (MCP) server in Python — then connect it to a Copilot Studio agent through Azure API Management. The reference example exposes Jira Cloud, but the patterns work for any enterprise system.

Start from the beginning →Jump to the Quickstart

Who this is for

This site is written for someone with limited Python experience and limited experience with agentic tools, who needs to understand the complete path from local development to a working Copilot Studio integration. Every chapter adds plain-English explanations, visuals, hands-on labs, and review checklists.

New to MCP

Start with Before You Begin for a jargon-free primer, then follow the chapters in order.

Want a quick win

Go straight to the Quickstart to run the server locally and pass the smoke test.

Shipping to production

Jump to Deployment, Security, and the Production Checklist.

What you'll build

A server that lets an AI agent safely use tools. In the reference implementation, the agent is a Copilot Studio agent and the external system is Jira Cloud. When a user asks “Find my open Jira issues”, the MCP server reads the signed-in user's delegated token, calls Jira as that user, trims the response, and returns a compact result.

The big picture. Every chapter zooms into one part of this flow.

The learning path

Nineteen chapters grouped into six stages. Work through them in order, or use the sidebar and search to jump around.

Get Started

Get Started

Before You Begin

A plain-English primer on the words and ideas you'll meet, so nothing feels like jargon later.
Get Started

Quickstart: Your First Success

Install prerequisites, open the project in VS Code, create a virtual environment, run the server, and pass the smoke test.

Understand

Understand

Core Concepts

MCP, tools, streamable HTTP, native MCP vs REST, delegated OAuth 3LO, APIM, Key Vault, and payload trimming.
Understand

Architecture & Request Lifecycle

How a user's question travels from Copilot Studio through APIM to the MCP server and Jira — and back.

Build

Build

Development Environment

Required software and VS Code extensions for Windows, plus how to open and run the reference project.
Build

Repository Anatomy & Python Basics

A guided tour of the project folders and the handful of Python ideas you need to read the code.
Build

Implementation Walkthrough

Layer by layer: server entry point, config, request-scoped tokens, middleware, auth, Jira client, trimming, tools.
Build

Local Development Workflow

Start the server, validate health, run the smoke test, use MCP Inspector, and run tests and lint.
Build

Build Your Own MCP Server

The repeatable 15-step method to design tools, auth, trimming, tests, and a connector for any system.

Integrate & Ship

Integrate & Ship

Copilot Studio & Atlassian OAuth

Create the Atlassian OAuth app, import the MCP connector, and add it to a Copilot Studio agent.
Integrate & Ship

Azure Deployment

Deploy with Bicep to App Service or Container Apps, front it with APIM, and configure the gateway policy.

Operate

Operate

Security Model

Non-negotiable rules, the identity flow, a threat model, and how to run a safe read-only deployment.
Operate

Testing Strategy

Existing unit tests, the tests to add before production, and a deploy-time smoke sequence.
Operate

CI/CD

The current GitHub Actions workflow and a recommended production pipeline with deployment gates.
Operate

Troubleshooting

Decision trees and a symptom-cause-fix table for the most common failures.
Operate

Extending the Server

Add new read and write tools safely, and port the whole pattern to another enterprise system.

Reference

Reference

Production Readiness Checklist

An interactive checklist across MCP, OAuth, security, payload, Azure, and CI/CD.
Reference

Glossary & Principles

Plain-English definitions of every key term, plus the final implementation principles.
How to use this site
  • Use the search box (press Ctrl + K) to find any concept fast.
  • Toggle dark / light mode with the button in the top bar.
  • Look for concept checks, labs, and end-of-chapter checklists — they make the ideas stick.