Securing AI Agents with MCP

LensesIO

Apr 29, 2026

Is your AI agent one misconfigured server away from a production data leak? In this deep dive, Jeremy from Lenses explores the critical security architecture of the Model Context Protocol (MCP) and how it’s evolving to protect the future of Agentic Engineering.

As we move past the era of 'vibe coding' into a more rigorous engineering discipline, AI agents now require direct access to your databases, APIs, and Kafka clusters. This video breaks down how to secure those connections using OAuth 2.1, the mandatory PKCE protocol, and the shift from Dynamic Client Registration (DCR) to Client ID Metadata Documents (CIMD).

Learn the step-by-step flow of how tools like Cursor and Claude authenticate with your infrastructure and what security layers are coming in 2026 to ensure your setup is production-ready.

Key topics covered:

What is Agentic Engineering and why MCP matters.
The 'M x N' registration problem in traditional OAuth.
How DCR (RFC 7591) allows programmatic client registration.
Why CIMD is becoming the new default for AI client identity.
The 5 future layers of MCP security: Fine-grained auth, machine identity, and more.

Stay ahead of the curve in AI infrastructure and cybersecurity.

Chapters:

00:00 - Why Your AI Agent is a Security Risk

00:36 - 2026: The Era of Agentic Engineering

01:25 - MCP: The Connective Tissue for AI Agents

03:32 - Why OAuth 2.1 is Mandatory for MCP

04:24 - The Scalability Problem in Traditional OAuth

06:41 - Understanding Dynamic Client Registration (DCR)

07:41 - Walkthrough: The Full OAuth 2.1 Flow Inside MCP

12:33 - Introducing CIMD: A Better Identity Model

13:27 - Step-by-Step CIMD Flow Explained

16:40 - DCR vs. CIMD: Key Differences & Trade-offs