How to Build Your Own Marketing Ops Assistant with MCP

AI agents are transforming Marketing Ops.

And as 2026 quickly approaches, those who not only use agents but also understand how they work and how to build them will have a massive head start.

It’s a very exciting time right now, where tools like MCP (Modern Context Protocol) now give Marketing Ops professionals the ability to create agents that work in their unique system and create real impact.

If you have no idea what MCP is, don’t worry. That’s exactly why we’re writing this. Because for MOPs folks, a good understanding of MCP can be a superpower.

A few weeks ago, RP’s own Lucas Gonçalves (VP of AI & Automation) had a great session at MOps-Apalooza 2025 centered around this exact topic of creating your own MOPs assistant using MCP. If you couldn’t attend in person or watch the session online, we’ve got you covered.

We’ve translated Lucas’s entire presentation into a streamlined guide that you can save and reference in the future.

Here’s what you’ll learn today:

What do we even mean by “AI Agent”?
What is MCP? Why should marketers care?
3 approaches to building your first MCP
A step-by-step walkthrough of building MCP in n8n
2 Marketo use cases with real implementations

Let’s get into it.

What Do We Mean by
“AI Agent”?

Before we dive into MCP, let’s gain a better understanding of what AI Agents are. The term gets thrown around quite a bit these days with wildly different meanings. Below, we’ll take a look at some common definitions and depth “levels” of AI Agent advancement. Then, we’ll highlight the definition that best suits our interpretation today.

Level 1: Point Solutions (The Google Cloud Definition) According to Google Cloud, “AI agents are software systems that use AI to pursue goals and complete tasks on behalf of users.” By this definition, almost anything qualifies as an AI agent. A custom GPT that answers specific questions, or even a fine-tuned model that classifies your database into personas based on job titles.
Level 2: Agentic Workflows (The Salesforce AgentForce Definition) This level involves structured processes where AI is embedded into workflows. Here, the AI follows certain steps and makes decisions along the way, but it’s still following a predetermined process. Think of this as a workflow that diagnoses the level of AI maturity in a company: the AI routes to option “A” or “B”, but doesn’t define the entire path itself.
Level 3: Autonomous Agents (The AWS Definition) AWS defines an agent as “a software program that can interact with its environment, collect data, and use that data to perform tasks.” This is where the AI itself will define the next steps rather than following a predetermined process. An autonomous Marketing Operations agent can perform tasks in your HubSpot, Marketo, or Salesforce according to a goal without you scripting every single action.

When we refer to “AI Agents” in this guide, assume we are focusing on “Level 3: Autonomous Agents” because this is where MCP really shines (and autonomous agents are also just cooler).

What is MCP? And Why Does it Matter for Marketers?

Put simply, MCP (Model Context Protocol) is a universal HTTP protocol that defines how AI models communicate with tools.

To understand why MCP is such a significant breakthrough, let’s use an analogy that folks in Marketing Ops will resonate with.

In MOPs, we are constantly juggling several platforms. And every platform has its own API. Nothing speaks the same language, and we often spend a ton of time just making systems communicate with each other.

This exact integration problem exists in the AI space too.

Without MCP, every major AI Model (be it ChatGPT, Claude, Gemini, etc.) needs to have custom integrations to work with other tools. And if you want to switch AI providers or if one of your tools updates its API, you need to rebuild all the integrations again.

But now with MCP, you only have to build the integration once, then any AI model can use it. Think of MCP kind of like a USB port: it’s one standard with broad compatibility. If you want to switch from ChatGPT to Claude, for example, your integrations through MCP will keep working. This is particularly important, seeing as the major AI models are constantly trading blows, outdoing each other every few months with new capabilities and improvements constantly being released.

(If only there were a similar magic protocol that MOPs folks could use that allowed Marketo, Salesforce, and every other platform to talk to each other despite updates and tool changes)

So, if MCP is not specifically a Marketing Ops technology, why should marketers care?

The reality is, AI Agents aren’t going anywhere. New ones are constantly being created, and existing ones are improving rapidly. Almost every major player in the MarTech space has committed to adding agents to their platforms. So if you, as a marketer, can learn a universal AI language like MCP, you can use it to build your own agents within your unique systems. It gives you far more agency (pun intended) and opens up a world of possibilities that amplify what your team is capable of.

3 Approaches to Building Your First MCP

Depending on your technical comfort level, there are three main paths to building your first MCP, which we’ll briefly outline below. Note: Each option is linked to a corresponding tutorial page.

Option 1: Claude Desktop (Best for Beginners)

If you’re just starting out and want to understand the basics, Claude Desktop offers an excellent introduction. Anthropic provides a step-by-step guide on creating a local MCP server that lets Claude interact with files on your system. It’s essentially copy-and-paste code that helps you see how an AI model starts interacting with your local environment.

Pros: It’s Easy to follow and great for learning.
Cons: It has basic functionality and it’s local only.

Option 2: FastMCP with Python (For Developers)

If you’re comfortable with Python, FastMCP is a library based on FastAPI that allows you to create servers you can host on AWS, Azure, or GCP. This approach gives you unlimited flexibility. You can make API requests, perform complex calculations, and implement any logic you can code.

Pros: This has maximum flexibility, it’s production-ready, and can be hosted for team access.
Cons: It requires Python knowledge and server management.

Option 3: n8n (The Sweet Spot for Most MOPs Pros)

n8n is an iPaaS (integration platform as a service) solution that offers a low-code/no-code approach to building MCPs. It’s like Zapier or Make.com, but with native MCP support. This is the approach we’ll focus on because it’s powerful enough for real-world use cases but doesn’t demand deep coding knowledge.

Pros: It’s a visual workflow builder with no heavy coding required and quick to deploy.
Cons: It comes with platform costs and it’s less flexible than custom code.

Let’s dive deep into this 3rd option below!

Building Your MCP in n8n: Step-by-Step

Now for the good stuff. Here is a step-by-step guide on creating an MCP in n8n.

Step 1: Set Up Your MCP Server Trigger

Create a new workflow in n8n.
Find the MCP trigger (it’s not in the obvious place: go to “Other Ways” and click “MCP Server Trigger”).
Customize your path (this is your endpoint URL, just like any API endpoint).
Add authentication if needed to secure your MCP.

Step 2: Add Your Tools

Tools are the actions your AI agent can perform. In n8n, you can add several types:

Code Tools: Write custom logic, like calculating averages or performing data transformations.
n8n Workflows: For complex, multi-step processes (like importing leads: create job → poll job → queue job → check status until complete).
HTTP Request Tools: Basic API calls to any endpoint.
Pre-built n8n Nodes: For platforms with native support (like HubSpot or Salesforce), these are plug-and-play. *Note on Marketo: Unfortunately, n8n doesn’t have pre-built Marketo nodes, so you’ll need to use HTTP requests and build the API calls yourself.

Step 3: Connect Your Agent

Now you need to wire up the AI agent that will use these tools:

Define the Trigger: Add a chat message trigger or your preferred input method.
Connect Your Agent: Choose your LLM (OpenAI, Claude, etc.).
Add the MCP Client: This connects your agent to the tools you defined.
Optional but Recommended: Add memory to give your agent context across actions.

Step 4: Add System Instructions

This is where you teach your agent how to behave. If you’re in Marketo, you’ll need to include:

Authentication details (Client ID, Client Secret, Munchkin ID)
System messages explaining how to perform tasks
Rules and guidelines for decision-making

Pro Tip: In production, send authentication data separately rather than including it in the agent instructions. But right now, for learning purposes, including it directly is fine.

What makes this architecture so powerful is that you can have multiple AI agents accessing the same MCP. You don’t rebuild this tool for each agent. The tools live within the MCP, and different agents access those tools as needed. This separation is what makes MCPs scalable!

Now, let’s jump into some real-world examples that show MCP and AI Agents in action.

Use Case 1: Automated Lead Merging in Marketo

In this example, we’ll be building an AI agent that can merge duplicate leads in Marketo. The process Marketers go through to merge leads requires several steps:

Finding the leads (you usually know the email, not the Lead ID)
Determining merge rules (which lead wins?)
Executing the merge

Building the MCP

For this use case, we need just three tools:

Tool 1: Get Access Token
- Before we can do anything, we need to authenticate with Marketo’s REST API.
  - Node Type: HTTP Request
  - Purpose: Obtain the bearer token for subsequent API calls

Tool 2: Find Lead by Email
- Since marketers work with emails, not Lead IDs, we need a way to look up leads.
  - Node Type: HTTP Request
  - Purpose: Query Marketo’s API to find leads matching an email address
  - Description: “Find leads in Marketo database by email address. Returns lead IDs and associated data.”
Tool 3: Merge Leads
- The actual merge operation once we have the Lead IDs.
  - Node Type: HTTP Request
  - Purpose: Execute the merge operation via Marketo’s merge leads endpoint
  - Description: “Merge two Marketo leads using their lead IDs. Preserves data according to specified rules.”

*Notice how each tool has a clear, detailed description? This is crucial. These descriptions tell the LLM what each tool does and when to use it. Without good descriptions, the AI will try to call tools without understanding their purpose, leading to errors and failed operations.

Building the HTTP Nodes

Creating HTTP request tools in n8n is straightforward. It’s like building an API call in any iPaaS:

Specify the endpoint URL
Set the HTTP method (GET, POST, etc.)
Add headers and parameters
The MCP difference: Add placeholders where the AI will dynamically insert data. For example, instead of hardcoding an email address, you’d use a placeholder like {{email}} that the AI fills in based on the user’s request.

Now here’s where it gets interesting. When you tell the agent to “merge these leads,” you haven’t specified the merge rules. Should it keep the oldest lead? The newest? The one with the most data?

This logic for this goes in the system message. Here’s an example of the rules you might include:

“When merging leads:
– Always prioritize the oldest lead as the winning record
– After merging, verify only one lead remains with the specified email”

You could also specify things like:

“Prioritize leads with a specific field value”
“Keep the lead with the most complete data”
“Preserve the lead that was created via form fill over imported leads”

It’s important to document your merge rules clearly so the AI can follow your organization’s standards.

Watching It Work

With all that set up, here’s what happens when you send the command: “Merge the two leads with the email address [email protected]”

AI Accesses MCP: The agent first calls the MCP to understand what tools are available
Gets Authentication: Calls the “Get Access Token” tool
Finds Leads: Calls “Find Lead by Email” with the provided email address
Performs Merge: Calls “Merge Leads” with the returned Lead IDs, applying the rules from the system message
Confirms Completion: Reports back to you what happened

The entire process takes just seconds (think of all that time you’d be spending manually navigating Marketo’s interface!)

Check Marketo and you’ll find just one lead remaining. All the merge rules you specified in the system message have been followed automatically.

Use Case 2: Cloning Programs and Updating Tokens

Our second example tackles another time-consuming task: creating new Marketo programs from templates. Every MOPs pro knows this pain:

Find the right template program
Clone it to the correct folder
Update all the tokens with new values
Update smart campaigns (which can’t be done via API)
Do final QA

This process can take 15-30 minutes per program. What if an AI agent could handle steps 1-3?

Building the MCP

This use case requires four tools:

Tool 1: Get Access Token
- Same as before: authentication is always step one.

Tool 2: Find Folder
- We need to locate where the new program should be created
  - Node Type: HTTP Request
  - Purpose: Search Marketo’s folder structure to find the destination
  - Description: “Find a Marketo folder by name. Returns folder ID and path information.”

Tool 3: Clone Program
- The core action—duplicating an existing program.
  - Node Type: HTTP Request
  - Purpose: Clone a Marketo program via the REST API
  - Description: “Clone a Marketo program to a specified folder. Creates an exact copy with a new name.”

Tool 4: Update Tokens
- Customize the new program with unique token values.
  - Node Type: HTTP Request
  - Purpose: Update program tokens via the API
  - Description: “Update token values in a Marketo program. Accepts token names and new values.”

Hardcoding vs. Dynamic Selection

In this example, the template program ID is hardcoded in the system message. But you could make it more sophisticated by:

Adding a “Find Program by Name” tool
Creating a “Create Folder” tool if the destination doesn’t exist
Adding validation tools to check token formats

Start simple with 3-4 tools and expand as needed. Even basic MCPs can save enormous amounts of time.

Now for the system message, once again. Here is what your system message might include:

“When cloning programs:

1. The template program ID is [12345]

2. Always clone to the folder matching the campaign name pattern

3. Update these standard tokens after cloning:

– {{my.CampaignName}}

– {{my.CampaignDate}}

– {{my.UTMCampaign}}

4. Follow our naming convention: [Region]-[Date]-[CampaignType]-[Description]

5. Do not activate any campaigns—human review is required”

Watching It Work

Command: “Create a program called MOPS-Apalooza-25 in the Events folder”

The agent:

Accesses MCP: Understands available tools
Gets Token: Authenticates with Marketo
Finds Folder: Locates the “Events” folder
Clones Program: Creates a copy of the template
Updates Tokens: Populates all tokens with appropriate values
Reports Success: Confirms the program is ready

Navigate to Marketo and you’ll now find:

New program in the correct folder
Properly named according to conventions
All tokens populated with correct values
Ready for a human to update smart campaigns and perform QA

If you have a well-templatized Marketo instance with everything properly tokenized, you can see the transformative potential here; Reduce program creation from 30 minutes to 30 seconds, eliminate manual errors in updates, and free your team to focus on strategy instead of repetitive tasks.

By now, you should have a solid foundational understanding of MCP, and how it transforms AI agents from abstract concepts into practical tools that solve real Marketing Ops problems.

As a marketer, speaking this unique language will be a major asset going forward as AI agents (and other unforeseen applications of AI) become increasingly more prominent and impactful in the MOPs space.

But having the technical know-how to build an MCP is only half the battle.

In Part 2 of this article, we’ll tackle the second half: how to prepare your systems and teams to work with AI Agents. How can you audit your tools for AI-readiness? How can you create solid documentation for your team? How can your team transition smoothly into new processes while avoiding common pitfalls?

We’ll answer all of these questions (and much more) in a new article coming soon titled: “How to Prepare Your Tools and Team for AI Agents”.

How to Build Your Own Marketing Ops Assistant with MCP

What Do We Mean by
“AI Agent”?

What is MCP? And Why Does it Matter for Marketers?

3 Approaches to Building Your First MCP