Skip to main content
Power of the MCE MCP
Control Marketing Cloud Engagement with Plain Language
Lukas Lunow
Trailblazer Community Virtual Session
Salesforce Marketer Group, Copenhagen, Denmark
Lukas Lunow
Lead Consultant,
Marketing Champion
NoA Ignite
Today’s Agenda
Key Takeaway
The MCP turns your AI from a tool
that plans campaigns into one that
actually builds them inside
Marketing Cloud.
And we’ll be honest about what to
watch out for.
What is MCP?
The open standard — think USB for AI
How MCP Connects to MCE
The flywheel: marketer, model, action
What It Enables
Journeys, Data Extensions, automations, content
Setup & Security
15-minute config — and the honest risk picture
Live Demo
MCP and Claude Code, live inside MCE
Best Practices
Getting started safely
Q & A
Bring your toughest questions
What is the MCP?
An open standard that lets any LLM take action in your systems — think of it as
the USB standard for AI.
Without the MCP
The AI is a consultant
Your LLM can generate content, write briefs,
and plan campaigns. But it stops at the plan.
Someone still has to open the UI and build
everything by hand.
Result: A plan on paper
With the MCP
The AI is an operator
The MCP gives your LLM the tools to act
directly inside Marketing Cloud — building
journeys, data extensions, and automations
from plain language. The plan comes to life.
Result: Real objects in MCE
How MCP Connects to MCE
The marketing MCP flywheel — turning strategy into executable reality.
You: The Architect
You bring the strategy and the
vision. You decide what
should happen.
Your LLM: The Blueprint
Claude, ChatGPT, or Gemini
turns your intent into logic,
copy, and plans.
The MCP: The Power
Tools
The bridge to MCE’s APIs. It
takes the plan and makes it
real.
The Result: Live in MCE
Journeys, data extensions,
and automations — built, not
just described.
What It Enables
Control Marketing Cloud in plain language - No code, no UI navigation required.
Journeys &
Automations
Build, publish, pause,
and manage journeys;
create and run
Automation Studio
workflows.
Data Extensions
Create, query, and
update DE schemas and
records; run complex
SQL.
Content
Draft emails, templates,
and SMS assets in
Content Builder
instantly.
Contacts & Sends
Retrieve and update
contacts and
subscription status;
manage email, SMS, and
push.
Reasoning: Probabilistic Execution: Deterministic
How the MCP Actually Works
Not a black box: the LLM picks the tool, the server maps it to MCE.
1. You Prompt
Create a data extension for
order history…
2.LLM Decides
Reads the tool manifest,
picks a tool, builds JSON
M
C
P
3. MCE Acts
The Data Extension gets
created via API call
Validates
and
translates
to API call
Renders
result back
to plain
English
What is an MCP tool?
User manual for your LLM
Tool name: sfmc_einstein_sto_activity
Full name: mcp__SFMC__sfmc_einstein_sto_activity
Description:
Build an Einstein Send Time Optimization (STO) activity JSON for insertion
into a Journey Builder journey.
## How It Works:
This tool generates the Einstein STO activity JSON and provides wiring
rules. Use the output with either:
- **sfmc_create_journey_builder_journey** - Include Einstein STO activity
when creating a new journey
- **sfmc_update_journey** - Add Einstein STO activity to an existing
journey
## What is Einstein STO?
Einstein Send Time Optimization analyzes each contact's engagement history
and determines the optimal time to send an email within a specified
look-ahead window (e.g., 24 hours). This increases open rates and
engagement.
## Parameters:
- target_activity_key: REQUIRED - The email activity key to insert STO
before (e.g., 'EMAILV2-1').
For existing journeys: Use sfmc_get_journey to find available email
activity keys.
For new journeys: Plan the email activity key upfront (e.g.,
'EMAILV2-1') and use the same key when building the email activity.
- duration: … [+997 chars]
Parameters:
● target_activity_key (required): string - REQUIRED: Email activity key
to insert STO before (e.g., 'EMAILV2-1'). Use sfmc_get_journey to find
available email activity keys.
● duration: integer - Look-ahead window in hours for STO (e.g., 24, 48,
72). Default: 24
Tools vs. Skills
→ Tools are the specific actions
the MCP can take, like “create a
data extension.” Salesforce
builds and maintains them.
→ Skills are your grounding files
and instructions — they tell the
LLM how to use those tools for
your business.
→ If Tools are the hammer
and nails, Skills are the
instruction manual tailored to
your team’s naming
conventions and processes.
→ Good Skills are what make
the agent get things right the
first time, instead of guessing.
Setting Up the MCP Server
Create the endpoint, point your client at it. About 15 minutes.
Step 1
Create Package
Installed Package + Scopes
In MCE Setup, go to Apps -
Installed Packages.
Add an API Integration
component and grant only the
permission scopes the agent
actually needs.
Step 2
Capture & Connect
Credentials + Endpoint
Note your Client ID and Tenant
ID.
Add the MCP server URL to your
client.
Pick the EU or US endpoint — this
is latency only, not data
residency.
Step 3
Authenticate
OAuth from Your Client
Authenticate your MCE instance
from inside Claude Code or
Gemini CLI.
No tokens are exposed.
Do this once, then you’re live.
MCP vs. Agentforce
Complementary, not competing. The MCP builds at scale to fill the funnel
Agentforce deploys.
MCP
Internal Build Tool
Your team builds faster behind
the scenes.
Campaigns, assets, journeys at
scale.
Operated by marketers, not
customers.
Agentforce Customer-
Facing
Live, real-time, two-way
conversations.
24/7 support and direct
conversion.
Speaks directly with your
customers.
Better Together
The MCP fills the funnel
Agentforce engages the funnel
One stack, two distinct jobs
Security: The Honest Picture
The MCP’s security model is about access governance.
Not data privacy at the LLM boundary. Know the difference.
Test in a sandbox first. Grant least-privilege scopes. Use plan mode before any write.
Oauth Scopes
Your floor
The agent acts as the
authenticated user. Its
powers are the intersection
of package scopes and your
permissions.
Prompt Injection
Likely
Data the agent reads
becomes prompt context.
Seeded content could
trigger unintended tool calls.
Ground and constrain.
No Native Log
Build it
MCP is stateless and doesn’t
log. MCE’s Audit Trail is a
security access log, not an
API activity log. Logging is
your job.
Data Leaves
Salesforce
Cleartext
With an external client, the
Einstein Trust Layer does not
apply. Your LLM provider’s
retention terms govern the
data.
Sandbox first
Test every workflow in a low-risk
environment (Sandbox BU)
before it touches production
data.
The agent can delete things.
Find that out in a sandbox.
Specific prompts
Your prompt is the instruction.
Vague intent produces vague or
wrong tool calls.
The LLM does the translating.
Garbage in, garbage executed.
Plan mode
Review what the agent intends
to do before it acts. Dry-run
destructive changes first.
Especially for schema changes
that ripple across DEs.
Least privilege
Grant only the scopes a use
case needs. Separate packages
for read-heavy vs. write work.
Grounding files
Skills tell the LLM how to use
the tools for your business —
your naming, your conventions.
This is what makes it get things
right the first time.
Verify output
AI-written SQL is often close
but not correct. Keep session
context clean and check results.
Trust, but verify — every query,
every change.
Best Practices: Getting Started Safely
Set your guardrails before you let an agent near production.
Demo Time
MCP + Claude Code
Live inside Marketing Cloud Engagement
Plain language, real results
What We’ll Build, Live
Create a Data Extension
From a csv file
Document
Help AI understand your platform
Run a query conversationally
And see what a vague prompt does
Inspect the result
Real objects, created from chat
Yes – skills are useful
They help you define your best practices
1. What to Trigger
Help select the right MCP tools for the job
2. Discover Before Acting
Use read/get tools to fetch current state before building or modifying
3. Validate Dependencies
Confirm referenced keys, IDs, and assets exist before composing JSON
4. Respect the SQL Dialect
SFMC uses a T-SQL subset — account for its specific constraints
5. Preview Before Execute
Generate and review output (JSON, queries) before committing to live assets
6. Avoid Surprises
Treat sends, publishes, and deletes as actions requiring explicit confirmation
Global/platform skill
MCE MCP mechanics that hold for any
org (discover-before-act, dialect
constraints, preview-before-execute).
Stable, rarely changes. This sets the
standard across all your implementations
Project/client skill
The org-specific layer (naming
conventions, compliance guardrails,
architectural patterns). Changes per
engagement – and should be very
lightweight, to avoid conflicts with the
global skill.
Use multiple skills
Worth knowing for all consultants out there
Resources & Q & A
MCE MCP Server
Setup & Tool Reference on Salesforce
Developers
https://developer.salesforce.com
/docs/marketing/mce-mcp
Useful Links
MCE MCP Setup Guide
https://developer.salesforce.com/docs/marketing/mce-
mcp/guide
MCP Tool Reference
https://developer.salesforce.com/docs/marketing/mce-
mcp/references/mce-mcp-tools/mce-mcp-tools.html
What the MCP Means for Marketing
https://www.salesforce.com/blog/marketing-mcp-server/
More from me
https://digitalmarketingoncloud.com
Bring your toughest questions
Let’s talk MCP — ask me anything
Harnessing the Power of MCP for Marketing Cloud Engagement Automation