Adobe Marketo Engage MCP module

Last update: October 2, 2026

The Adobe Marketo Engage MCP module allows you to send a natural-language prompt to Adobe Marketo Engage’s MCP (Model Context Protocol) server, using an AI model to interpret the request and call Marketo’s own tools to fulfill it. Unlike a traditional Marketo connector where each module does one fixed action, such as “Create a Lead”, this connector has a single module that accepts an open-ended instruction in plain English and lets the AI decide which Marketo operations are needed to satisfy it.

This connector is for Marketo Engage’s own MCP server specifically

To connect to MCPs for other applications, see see Add an AI prompt to your scenario.

Access requirements

Expand to view access requirements for the functionality in this article.

table 0-row-2 1-row-2 2-row-2 3-row-2 layout-auto html-authored no-header
Adobe Workfront package	Any Adobe Workfront Workflow package and any Adobe Workfront Automation and Integration package Workfront Ultimate Workfront Prime and Select packages, with an additional purchase of Workfront Fusion.
Adobe Workfront licenses	Standard Work or higher
Adobe Workfront Fusion license	Operation-based: Available to organizations with operation-based licenses Connector-based (legacy): Workfront Fusion for Work Automation and Integration
Product	If your organization has a Select or Prime Workfront package that does not include Workfront Automation and Integration, your organization must purchase Adobe Workfront Fusion.

For more detail about the information in this table, see Access requirements in documentation.

For information on Adobe Workfront Fusion licenses, see Adobe Workfront Fusion licenses.

Prerequisites

You must have an Adobe Marketo Engage account and a valid Marketo instance.

Connect Adobe Marketo Engage MCP to Workfront Fusion connect-adobe-marketo-engage-mcp-to-workfront-fusion

You can create a connection to your Marketo instance directly from inside the Adobe Marketo Engage MCP module.

In the Adobe Marketo Engage MCP module, click Add next to the Connection field.

Fill in the following fields:

table 0-row-2 1-row-2 2-row-2 3-row-2 4-row-2 5-row-2 layout-auto html-authored no-header
Connection name	Enter a name for the new connection.
Environment	Select whether you are connecting to a production or non-production environment.
Type	Select whether you are connecting to a service account or a personal account.
Client ID	Enter the Client ID for your Marketo REST API service, as created in Marketo LaunchPoint.
Client Secret	Enter the Client Secret for your Marketo REST API service, as created in Marketo LaunchPoint.
Munchkin ID	Enter your Marketo instance's Munchkin ID (for example, `123-ABC-456`). The Munchkin ID is shown in Marketo under Admin → Munchkin.

Click Continue to create the connection and return to the module.

IMPORTANT

Use a dedicated API-only Marketo user with the minimum role and permissions required for the scenario, rather than reusing an administrator account.
Creating the connection does not validate the credentials. Fusion saves them without a test call, so the connection can appear to be created successfully even if a value is wrong or mistyped. If a credential is incorrect, the failure usually appears later when the module first tries to reach Marketo or when the tool lists fail to load.

The module: “Process a user prompt”

This is the only module the connector provides. A scenario uses it by supplying:

Connection — the Marketo connection created above.
Enter your prompt — the instruction, in plain English (for example, “find every lead added to the Spring Webinar list in the last week and tell me which ones have no company name set”).
Tools (optional) — described below. These fields only appear once a connection is selected.
LLM key (optional, advanced) — described below.

It returns the AI’s final answer as text, plus a full audit trail of what happened while producing that answer.

Adobe Marketo Engage MCP module and its fields

Process a user prompt

This action module sends a plain-English instruction to Adobe Marketo Engage’s MCP server and returns the AI’s response.

LLM key (Optional, advanced)

By default, this module processes your prompt using Adobe's own AI service, and you do not need to select a key.

To use your own AI provider instead, select an existing LLM key, or create a new one by clicking Add and entering the following information:

Key name: Enter a name for the new key.
LLM: Select the large language model that this key is associated with. Supported providers are OpenAI, Anthropic Claude, and Amazon Bedrock.
Key: Enter or map your API key for the selected provider.
Model: Select the LLM model that the key will use.
Other fields: Enter values for any other fields that your LLM requires.

Connection

For instructions about connecting your Marketo account to Workfront Fusion, see Connect Adobe Marketo Engage MCP to Workfront Fusion in this article.

User prompt

Enter or map the instruction, in plain English, that you want the AI to carry out.

Example: Find all leads added to the Spring Webinar list in the last 7 days and summarize which industries are most common.

Module output

The output is a single bundle containing the following:

Response: The AI’s final answer, as text. You can map this data into subsequent modules.
Audit Trail: The detailed record of the run, including a session ID, the original prompt, start and end times, total duration, overall status, the final response, and a Tool Calls list. Each tool call entry records which Marketo tool ran, its arguments, its output, its start and end time and duration, whether it succeeded, and its order in the sequence.
Summary: The same run condensed to counts: total tool calls, successful calls, failed calls, processing time, and status.

AI Models

By default, the module uses Adobe’s own managed AI service automatically, with no key or credentials to enter.

You can instead select a specific LLM key to use OpenAI, Anthropic Claude, or Amazon Bedrock, if your organization has an account with one of these.

Choosing which Marketo actions the AI is allowed to take

After a connection is selected, the module asks the Marketo MCP server what tools it offers and presents them as multi-select lists, each showing how many tools it contains:

Read-only tools: Actions that only look something up and never change anything, such as finding a lead, listing campaign members, or reading a program’s details.
Write/delete tools: Actions that change something, such as creating or updating a lead, adding someone to a list, activating a campaign, or approving or sending an email.
Other tools: A third list that appears only if the Marketo server offers tools it has not labeled as read-only or not. These are shown separately rather than being assumed safe or unsafe. If the server labels everything, this list does not appear.

If no tools are selected, the AI can use them all. You can restrict a list to specific actions. For example, selecting only 2 specific “write” actions while leaving “read-only” alone means the AI can look up whatever it needs freely, but can only make those 2 specific kinds of changes. Leaving a list empty means all actions in that category are allowed. Restricting the AI requires actively choosing which specific actions to allow in that category. This way you can ensure that the AI won’t take an unexpected destructive action against live marketing data, while still letting it freely gather information.

Because the lists are read live from the Marketo server, the exact tools shown can change as Adobe updates that server.

No persistent conversation history

Each run of this module is a single, self-contained execution. The AI cannot ask a follow-up question and wait for a reply. It must instead make its best judgment and produce a complete, final answer in one pass. If a request is ambiguous, the AI will make a reasonable assumption, state that assumption as part of its answer, and proceed. It will not stop and ask the user to clarify, because there’s no way for it to receive a reply within a single run.

The AI is also instructed to verify facts with a tool call rather than rely on memory, because Marketo data may have changed since the previous run.

The AI only takes a write, update, or delete action when the prompt actually asked for one. It won’t take an action that wasn’t requested, including activating or deactivating campaigns, creating or deleting leads and lists, and approving or sending emails, even in the same run where it’s doing something else the user did ask for.

Because each run is independent, the AI has no memory of a previous run by itself. A scenario that wants a multi-turn, chat-like experienc emust explicitly provide that history as part of the new prompt, such as by storing the previous question and answer in Fusion’s Data Store, or passed between modules, and including it as text at the start of the new prompt, followed by the new question. There is no session or conversation ID that automatically remembers prior runs.

Example prompts

You can use prompts such as the following:

List the leads that joined the ‘Q3 Product Launch’ program in the last 7 days and summarize which industries they’re from.
Check whether the ‘Welcome Series’ smart campaign is currently active, and tell me how many people are in it.
Find the form used on our pricing page and tell me which fields are marked required.
Add the lead with email jane@example.com to the ‘VIP Customers’ static list.
Summarize the performance of every email in the ‘Spring Newsletter’ program.

recommendation-more-help

workfront-fusion-help-workfront-fusion