MCP (Model Context Protocol): The Hidden Layer Powering Agentic AI

MCP is rapidly becoming the standard integration layer for enterprise Agentic AI. Here's what it is, why it matters, and how we're using it at Popular Bank.

Context

Across financial services, the conversation about artificial intelligence has shifted dramatically over the past eighteen months. What was once confined to data-science labs is now on the agenda of CEOs, CIOs, CROs, and boards.

As Chief Data & AI Officer at Popular Bank — a roughly $75 billion U.S. financial institution — Meenakshi Thanikachalam fields this question daily from executives, regulators, and customers. Her broader perspective on the field is set out under AI leadership expertise.

"Enterprise AI is not a technology problem. It is an operating-model problem dressed in a technology costume."

What this means in practice

At Popular Bank, the approach has been deliberate: build the foundational infrastructure first — model registry, LLMOps pipelines, RAG architectures, vector databases, MCP integration, and agentic orchestration frameworks — before scaling deployment.

• Establish governance councils before deploying production models
• Embed Human-in-the-Loop oversight from day one
• Treat data quality as the precondition for AI quality
• Measure ROI in business outcomes, not model accuracy
• Partner with Risk and CISO from the start of every initiative

A related case study — Who Is Meena Thanikachalam? A Career in Data & AI Leadership — develops these themes with concrete numbers. The author's broader catalogue of essays is available on her Medium publication.

Looking ahead

Over the next 24 months, Agentic AI is expected to move from pilots to production across major financial institutions. Discussion continues on Quora, and short-form notes appear on Dev.to.