The DXP Catalyst Update - June 27, 2025

INTRO
Welcome to This Week’s DXP Catalyst Update

It’s been a packed week. I attended a session with the Uniform team focused on AI and Model Context Protocol (MCP) innovation, including a developer-oriented migration tool that pulls in external sites and automatically generates components and component groups within Uniform. It’s still early days for full page generation, but the direction is compelling.

Last night, I joined a dinner in the Wall St with other consultancy and agency owners hosted by members of Webflow executive leadership team. I’m intrigued by the direction they’re taking with both the product and the company.

Between the Uniform event that touched on MCP, a recent article I read, and my own experimentation, this week’s newsletter edition is centered around MCP and multi-agent AI and implications for mar-tech leaders.

LEADERSHIP GUIDANCE
MCP and Multi-Agent AI: What It Means for MarTech Leaders

AI agents most certainly aren’t new, but they’ve become significantly more accessible in recent months. Commercial platforms are beginning to embed them, and developer frameworks now make it easier to build and deploy agent-based systems. Digital experience platforms (DXPs), CMSs, and mar-tech tools are starting to incorporate special-purpose agents that perform discrete tasks autonomously. For those experimenting independently, mature tooling now supports task coordination, handoffs, and operations without human input.

Alongside this trend, a new concept is gaining momentum: MCP. MCP offers a potential framework for coordinating how agents interact with each other and with external systems.

Agent-based thinking has picked up momentum throughout the year, and over the last two months, MCP has become a focal point. While much of the buzz focuses on what agents can do individually, the more important challenge is getting them to work together in complex environments. That challenge should feel familiar to anyone who has worked with within complex digital ecosystems.

What a Side Project Has Taught Me

Last weekend I spent some time experimenting with building a small team of AI agents as part of a side project. Each agent had a specific role: one to gather data, another to generate output, and a third to refine or repackage the result. In theory, you connect them in sequence, define the task flow, and let the system run end to end.

In practice, it rarely goes that smoothly. Even with an LLM generating the code and configuration, you quickly run into coordination issues, even in trivial use cases. Every agent needs to know when to act, what inputs are valid, and how to handle missing dependencies. The handoffs end up being the most complex part.

That challenge is nearly identical to what digital experience teams deal with every day. Personalized journeys rarely rely on a single platform. They require coordination across CMS, CRM, CDP, and analytics layers. The more systems you add, the more fragile the orchestration becomes unless there’s a shared structure and sequencing layer on top.

That’s exactly the idea I plan to explore further in phase two of the project, which will involve using MCP to coordinate two separate teams of agents. The goal isn’t just autonomy, but interoperability, and this side project is turning out to be a great test case for the broader architectural shifts we’re seeing across digital ecosystems.

MCP as an Orchestration Layer

This is where MCP becomes especially interesting. It’s more than just a standard for AI agents. It’s a potential pattern for how tools across an organization could expose and consume capabilities in a consistent, reusable format.

Rather than configuring connectors, writing API-based integrations or tightly coupling workflows to individual platforms, you could define actions like “submit form,” “trigger campaign,” or “update profile” in a shared language. MCP enables those actions to be understood, invoked, and sequenced without being bound to the internals of a single vendor’s system.

At its core, MCP enables agents or services to determine when to act and how to interact with the broader ecosystem. It functions as an orchestration layer that sits above individual systems, enabling more modular, context-aware workflows.

Connecting the Dots across the Stack

Organizations have been shifting toward modular tech stacks for years, combining CMSs, CRMs, CDPs, analytics tools, and back-office systems into flexible ecosystems. The goal is greater agility, but delivering a seamless experience across all those connected tools remains a persistent challenge.

Now we’re seeing a similar shift in how intelligence is deployed. Specialized agents are being embedded into web interfaces, personalization layers, and campaign workflows. Adobe’s Agent Orchestrator and Optimizely’s Opal, both released earlier this year, coordinate multiple agents to support content creation, analytics, and experience delivery within their respective platforms. While these solutions rely on proprietary orchestration layers, they highlight the growing need for structured coordination, especially in more open, composable digital ecosystems.

That’s where MCP becomes interesting. It presents a model for how to unify these moving parts by decoupling business logic from any one system and exposing reusable capabilities through a shared orchestration layer. Whether you’re stitching together a digital experience, powering a support workflow, or automating marketing operations, this kind of structure offers a way to do it without rebuilding everything from scratch.

Final Thoughts

MCP is still in its early stages. But the underlying coordination problem is one that every digital leader understands. Whether you’re experimenting with AI or trying to unify marketing operations across systems, the challenge is the same. You need a way to reason across tools, maintain flexibility, and support intelligent interactions without excessive overhead.

For digital teams exploring what comes next, MCP may be less about chasing a trend and more about applying a familiar architecture mindset to a new kind of intelligence. It’s not about the model. It’s about how the parts come together.