The Adobe Marketo Engage MCP Server, Explained: What It Is, What It Can Do, and How to Get Started
The Adobe Marketo Engage MCP server is a model context protocol server that lets LLMs access Marketo Engage using plain language instead of hand-formatted API calls to unlock agentic capabilities. Adobe’s description? “It as “a bridge between your AI assistant and Marketo.”
The mechanics are simple and like any other MCP server: When you ask a question and give your LLM a prompt, for example, “which smart campaigns are active right now?”, your AI assistant decides which Marketo Engage operation answers it, calls the server, and the server executes (using the context provided by the MCP server) and the answer comes back into your AI tool, in your conversation.
But of course, the follow up question is…
What Is the Model Context Protocol (MCP)?
The Model Context Protocol, or MCP, is an open standard for connecting AI models to external data sources and tools. It exists to solve a combinatorial problem: without a shared standard, every AI tool needs a bespoke integration with every platform it wants to reach. With MCP, a platform publishes one server and any compliant AI client can connect to it.
The comparison people reach for is a universal port like a USB, one connector shape, many devices. A more precise description, if you work with Marketo’s APIs already, it is an API with context attached that uses natural language to execute workflows rather than structured API calls.
A conventional REST call requires you to know the endpoint, the payload structure and the authentication dance. An MCP server publishes its operations with enough descriptive context that a language model can work out which one to use from a plain-language request, then call it correctly.
How is the MCP server different from Marketo AI?
This is the most common point of confusion, and it is worth being precise about, because the two capabilities solve different problems and are provisioned separately.
Marketo AI is built into Marketo Engage. It provides a conversational interface inside the product and a set of purpose-built agents for specific marketing ops jobs, importing and cleaning a lead list, validating a program before launch, generating a smart list from a description. Adobe supplies the models, the interface and the guardrails.
The MCP server works in the opposite direction. It adds nothing to the Marketo interface. Instead it opens your instance to the AI tool your organization has already standardized on, and you write your own prompts.
They are complementary rather than competing, and in practice most teams will end up using both. If you want a defined job done consistently by anyone on the team, Marketo AI is the better fit. If you want to ask something nobody built a feature for, or pull Marketo data into a conversation that also touches your CRM, that is the MCP server.
What can the Marketo Engage MCP server do?
Adobe documents more than 100 operations, grouped into seven areas:
| Forms | Browse, create, clone and approve forms; manage fields and field visibility rules |
|---|---|
| Smart Campaigns | Create campaigns, configure filters, add flow steps, activate and deactivate |
| Leads + Smart Lists | Find leads, create and update records, manage list membership |
| Programs | Create, clone and tag programs; browse by type, channel or date |
| Emails | Browse, create from templates, update content sections |
| Structure | Browse folders, channels, tag types and activity types |
| Bulk Operations | Export lead data and check job status |
MCP Use Case #1: asking questions
This is where the server is strongest and where anyone should start:
- Which campaigns are running
- What is this person’s history and why did they score the way they did
- Which forms exist, and what is using them
- These are read-only, cheap, and impossible to break anything with, and they cover a large share of the interruptions a marketing ops team fields in a week.
Adobe’s own example prompts sit in this tier: “Show me all approved forms,” “What smart campaigns are active right now?”, “Find the lead with email [email protected],” “Show me all draft emails.”
MCP Use Case #2: auditing and tracing
Following dependencies, finding orphaned assets, working out what a change would break. Marketo is dependency-heavy, a form cannot be unapproved while a smart campaign references it, and tracing those relationships by hand is slow enough that doing it conversationally is a genuine improvement.
Expect to reformulate prompts here, because the answer depends on what the underlying API actually exposes.
MCP Use Case #3: creating and changing
Cloning a program, building a smart list, drafting an email from a template. This works, and Adobe’s documented example, “Clone the Q4 Webinar program into the 2026 Events folder”, is a good illustration of where it fits.
But this tier is also where you will meet the most friction, because Marketo’s creation rules are strict: emails must live inside a program, templates must be approved, and asset names collide across a workspace.
The practical guidance: use the MCP server for cloning existing structures and for querying, and stay in the Marketo interface or use Marketo AI for building something genuinely new. Cloning a working program is faster conversationally. Constructing one from nothing usually is not.
What the MCP Server won’t
Four limits worth knowing before you invest time.
- No reporting or analytics. Campaign performance, email statistics and attribution reporting are not in the documented operation set. If that is your goal, this is the wrong tool.
- It does not add permission granularity. Role-based access is inherited from the Marketo API user you configure. Whatever that user can do, anyone using the server can do. There is no per-category scoping layer on top of it, which is why Adobe recommends a dedicated API user holding only the permissions actually required.
- It consumes your Marketo API quota. Every prompt becomes one or more real REST API calls, and a single natural-language request can fan out into several. Ordinary interactive use is unlikely to trouble a standard daily limit; anything automated or scheduled deserves a calculation first. Hitting the ceiling returns a “too many requests” error.
- It is a beta on an emerging standard. Adobe’s own documentation carries a notice that MCP “is an emerging open-source standard and may present security or reliability risks.” Treat it accordingly: sandbox first if you have one, and do not build a business-critical process on it yet.
What does it cost?
There are two meters running, and they belong to different vendors.
On the Adobe side, you spend Marketo API calls against your instance’s existing quota. There is no separate charge for the MCP server itself. In practice a handful of people asking a handful of questions a day is a rounding error against a typical daily limit, but because one prompt can trigger several calls, automated use needs modeling rather than assuming.
On the AI tool side, you spend tokens, just as you would using your LLM for anything else, you have to keep in mind what your request requires from a token spend standpoint.
What do you need to connect it?
Five prerequisites:
- A Marketo Engage instance with REST API access enabled.
- Admin access to create API credentials in LaunchPoint, which yields a client ID and client secret. Create a dedicated API user for this rather than reusing one wired into another integration.
- Your Munchkin account ID, from Admin > Munchkin, in the format XXX-XXX-XXX.
- Beta access, requested via Adobe’s signup form with that Munchkin ID.
- A supported AI client.
Adobe documents configuration for Claude Desktop, Claude Code, Cursor, OpenAI Codex, VS Code with GitHub Copilot, and Glean, and notes that “any MCP client that supports remote servers over streamable HTTP transport may connect to it.” Beta participants have also connected through the Gemini CLI, which is worth knowing if Google is your organisation’s standard, though it is not among the documented configurations.
One difference is worth flagging because it catches nearly everyone. Most clients connect directly to Adobe’s hosted endpoint over HTTP, with authentication in request headers. Claude Desktop does not — it needs a small local bridge (a downloadable package plus Node.js v18 or later on your machine) because Desktop does not connect to remote MCP servers natively. If you are using Claude Desktop and the documentation looks like it is describing a different product, that is why.
Who should be using this right now?
A small number of people, deliberately chosen, get real value from this today.
Marketing ops practitioners get the clearest return, and mostly not from building things. The gain is in answering the questions that arrive by Slack all day, why is this person in this campaign, what is using this form, what changed last week, without stopping to dig.
Another interesting audience is people who never open Marketo at all. A BDR, an account executive or a customer success manager can ask about a record and get a coherent answer without learning the platform. That is a genuinely new capability rather than a faster version of an old one, and it is probably where the durable value sits.
What this is not yet ready for is a broad rollout. With permissions inherited wholesale from a single API user, no usage metering exposed, and an explicit beta caveat on reliability, handing it to a large team would be premature. The right shape today is a small group who know Marketo well, working through a dedicated least-privilege API user, ideally in a sandbox, learning what is worth systematizing before anything gets built on top.
Three prompts to start with
If you have it connected and you are not sure what to type, these three are cheap, read-only, and demonstrate the range in about five minutes.
- “In Marketo, list all the smart campaigns that are currently active, grouped by the folder they live in.”, Establishes that the connection works and returns something you can immediately sanity-check.
- “In Marketo, show me all of the activity for [an internal email address] and tell me what led to them becoming a prospect worth reaching out to.” The single most useful thing the server does, and the one most likely to make the value obvious.
- “In Marketo, list the forms in this workspace and tell me which assets depend on each one.” Shows dependency tracing, and gives you a genuine audit artifact as a by-product.
Note the pattern: each one names Marketo explicitly, scopes to a specific object, and asks for an interpretation rather than a data dump. That shape is most of what separates a useful answer from an expensive one.
Frequently Asked Questions
Do I need coding skills to use the Marketo Engage MCP server? No. The server translates plain-language prompts into Marketo REST API calls, so the skill that matters is knowing what to ask and how to scope it to a specific object, like a program, list, or form. Marketing ops practitioners without a developer background can use it directly through a supported AI client.
Which AI tools currently support the Marketo Engage MCP server? Adobe documents configuration for Claude Desktop, Claude Code, Cursor, OpenAI Codex, VS Code with GitHub Copilot, and Glean, and states that any MCP-compatible client should work. Most clients connect directly to Adobe’s hosted endpoint over HTTP.
How is the MCP server different from asking a question inside Marketo Engage itself? Marketo AI lives inside the product and runs on Adobe’s models through purpose-built agents for specific jobs. The MCP server instead opens your Marketo instance to whatever AI tool you already use, running on your own tokens rather than Adobe’s compute.
Will the MCP server slow down or interfere with my existing Marketo workflows? No new interface appears inside Marketo, and nothing is installed on infrastructure you manage. The only operational impact is that each prompt consumes calls against your existing Marketo API quota, which can matter for teams already close to their limit.
Is it safe to give an AI tool access to Marketo data through this server? Adobe states the server runs no AI models, does not train on customer data, and does not store credentials, supplying them per request instead. Access is scoped to whatever permissions the connected Marketo API user holds, so the practical review should focus on that user’s permissions and your AI tool’s own data handling.
Does the MCP server use AI?
No. Adobe states the server “does not use, host, or run any AI or machine learning models.” It translates requests into Marketo REST API calls. All model processing happens in the AI tool you connect.
Does Adobe train AI models on my Marketo data through the MCP server?
No. The documentation states the server does not train on customer data and does not generate predictions. It also does not store credentials or session state between requests.
Can I use ChatGPT, Gemini or Microsoft Copilot with the Marketo MCP server?
Adobe documents Claude Desktop, Claude Code, Cursor, OpenAI Codex, VS Code with GitHub Copilot, and Glean, and states that any MCP client supporting remote servers over streamable HTTP may connect. Beta participants have connected Gemini through its CLI. Check current documentation for your specific client before committing.
What is the difference between the MCP server and Marketo AI?
Marketo AI is a conversational interface and set of purpose-built agents inside Marketo Engage, running on Adobe’s models. The MCP server connects your own external AI tool to your Marketo instance and runs no models itself. They are separate capabilities with separate setup requirements.
Does the MCP server count against my Marketo API limits?
Yes. Every operation is a real REST API call against your instance’s quota, and one natural-language prompt can trigger several calls. Exceeding the limit returns a “too many requests” error.
Can I control who can do what through the MCP server?
Only through the Marketo API user it is configured with. Role-based access is inherited from that user’s permissions, and the server does not add finer-grained per-operation controls. Use a dedicated API user with the minimum permissions required.
Is the Marketo MCP server generally available?
At the time of writing it is in limited beta. Access requires completing Adobe’s signup form with your Munchkin account ID. Adobe also notes that MCP is an emerging standard that may present security or reliability risks.
Key Takeaways
- The Marketo Engage MCP server lets AI tools like Claude or Cursor read from and write to Marketo using plain-language prompts instead of API calls.
- It differs from Marketo AI: Marketo AI runs inside Marketo on Adobe’s models, while the MCP server runs through your own AI tool and token budget.
- It supports over 100 operations across forms, campaigns, leads, programs, emails, and bulk exports, but has no reporting or analytics capability.
- Adobe states the server runs no AI models, stores no credentials, and does not train on customer data, though access permissions are inherited wholesale from a single Marketo API user.
- It is currently in limited beta and best suited to a small group of marketing ops practitioners and cross-functional users asking read-only questions.
Where to go next
If you want to see this working rather than read about it, we recorded a walkthrough that takes a single record and traces its entire history in one prompt — the second of the three prompts above, run end to end against a live instance. And if you are trying to decide whether the MCP server or Marketo AI is the right fit for a specific workflow, that is a conversation we have most weeks; we are happy to have it.
About the author. Akande Davis is VP Operations at GNW Consulting, an Adobe-recognized Marketo Engage Champion, and has been running the Marketo Engage MCP server against a live production instance since the alpha program.
Sources. Adobe Experience League, “MCP Server for Marketo Engage.” Adobe Business blog, “Agentic AI is redefining the user experience for marketing ops.” Observations from Adobe’s Marketo MCP alpha and closed beta programs, April–June 2026.
Last updated: August 7, 2026 – this article covers a product in active beta and is reviewed monthly.
