Building MCPs with Entra: Design Choices for Enterprise Agents

A 60-minute technical session on building enterprise-grade MCP (Model Context Protocol) servers with Microsoft Entra ID – covering identity, consent, authorization, auditability, and the new Agent ID principal type.

Presented by Merill Fernando

Session Recording

Slides

View the slides (PDF) | View the slides (HTML) | Slide source (Marp)

What You’ll Learn

1. How MCP auth works – PRM, ASM, CIMD (SEP-991), DCR, Resource Indicators, and how the spec evolved
2. Where Entra fits (and doesn’t) – what’s supported, what’s not (CIMD, DCR), and why
3. Auth model first – stdio vs remote auth options, decision trees, choose before you build
4. Reference architecture – Microsoft Graph MCP Server as the exemplar, then apply it to your API
5. Entra Agent ID – the new ServiceIdentity principal for AI agents
6. Which Azure platform – Functions vs App Service vs APIM vs Container Apps
7. How to audit and scale – Graph Activity Logs, Sentinel, multi-tenant, caching
8. What to watch out for – 36 documented pitfalls from real deployments

Repo Structure

mcpentra/
├── slides/          # Marp slide deck (md, pdf, pptx, html)
├── docs/            # 17-document knowledge base with deep-dive content
└── demo/            # Contoso Directory MCP Server (TypeScript demo project)

Knowledge Base

The docs/ folder contains the full knowledge base behind the presentation:

Key Standards