Next AI CMO lunch: April 28, 2026 · San FranciscoAI CMO lunch · Apr 28 · SFReserve your seat →
All posts

News

Adobe's Marketo MCP Is Coming. Here Are the Concepts You Need to Know.

Apr 20, 2026 · alysha khan

Adobe Marketo MCP explainer — video thumbnail

Adobe is announcing a Marketo MCP at Summit next week. If you've spent the last few months nodding along in meetings when someone says "MCP" like you know exactly what that is, this is your cheat sheet.

What an MCP actually is

MCP stands for Model Context Protocol, but the acronym matters less than what it does. An MCP is how a language model takes action in a real system. You plug it into your model, and suddenly the model can do things in that system: read data, make changes, run processes.

Think of it like an API, but built for models instead of developers. APIs expose functionality to engineers who write code against them. MCPs expose functionality to LLMs, along with the documentation the model needs to figure out how to use each action on its own.

When Adobe ships the Marketo MCP, it means you can connect your model of choice (Claude, ChatGPT, Gemini, whatever) and have it actually work inside Marketo. Not summarize a screenshot of Marketo. Not draft copy you then paste into Marketo. Do the work, inside the instance.

The flavors of Claude, and when each one matters

Before you can use an MCP, you need to know which version of Claude you're plugging it into. Here's the lineup:

Claude Chat is the web or mobile app most people know. You type, Claude responds. Great for questions, drafting, analysis, brainstorming. It doesn't have access to your files or systems unless you paste content in directly.

Claude Code is a command-line agent built specifically for coding work. It can handle complex, multi-step engineering tasks. It got so capable that all sorts of people outside of engineering started running wild with it for non-coding work, which is what led Anthropic to build…

Claude Cowork, a desktop tool where you hand Claude access to a folder and it autonomously reads, edits, and creates files to complete full tasks. Not just asking a question but delegating a whole project. Same underlying capability as Claude Code, but designed for people who don't live in a terminal.

The MCP is the capability layer that plugs into any of these. You pick the Claude that matches how hands-on you want to be, then connect the MCP so the model can reach into Marketo from there.

Where Claude Skills fit, and why people confuse them with MCPs

Claude Skills are reusable instruction packs that Claude automatically loads when they're relevant to your task. They often contain documentation, procedures, and sometimes small scripts that teach the model how to do something specific. When Claude uses a skill, it's essentially reading the playbook for that task and then applying it.

Skills and MCPs both feel like "plugging the model into something," but they do different jobs:

  • A skill teaches the model how something is done.
  • An MCP gives the model the ability to do it in a real system.

For most real MOps work, you need both. The skill tells the model how to build a webinar program the way your company builds webinar programs. The MCP lets the model actually create the program, the tokens, the smart campaigns, and the assets inside Marketo. Without the skill, the model has capability but no expertise. Without the MCP, it has expertise but no hands.

The analogy that makes it click

The skill is the cookbook. The MCP is the kitchen. The LLM is the chef.

The chef picks up the cookbook, flips to the recipe for what you asked for, and then uses the kitchen to actually cook it. The cookbook alone doesn't feed anyone. The kitchen without a recipe gets you an inconsistent mess. Put all three together and you get dinner.

Why this matters for MOps

Every vendor is racing to ship an MCP right now, and Marketo joining the list is a big deal for anyone running a Marketo instance. The work that used to require either a practitioner's hands on keys or a custom API integration is about to be reachable by models directly, with the right skills and the right guardrails in place.

The practitioners who figure out how to design good skills, connect the right MCPs, and build repeatable workflows on top of them are going to move a lot faster than the ones waiting to see how it shakes out. If you're in MOps and you've been circling AI adoption without a clear entry point, the Marketo MCP is a real one.

Ready to hire Mary?

Less reading. More shipping.

Book a Demo