Salesforce Marketing Cloud MCP Server: A Practical Guide for MOps Teams

For years, the ceiling on what a marketing operations team could execute was set by three constraints: technical resources, content production capacity, and data access. The ideas were never the bottleneck. It was mostly the tools. 

The Salesforce Marketing Cloud MCP Server, shipped as part of the Summer ’26 release, is a direct attack on that bottleneck. It lets AI assistants like Claude and Gemini CLI operate against your Marketing Cloud account through natural language. There’s no middleware, no custom Apex, no third-party connectors. You describe what you need. The AI executes it.

This post covers what the SFMC MCP Server is, what it can actually do, how to set it up, and whether your team is positioned to get value from it right now.

What Is MCP, and Why Should SFMC Users Care?

Understanding Model Context Protocol 

MCP (Model Context Protocol) is an open standard, originally published by Anthropic, that allows AI assistants to connect directly to external systems. Before MCP existed, every AI integration required custom connectors with bespoke code built for one tool, one use case, one team. MCP replaces that fragmentation with a single, standardized protocol. This means one server can be used across any compatible AI client.

Think of it as a universal interpreter. Your AI assistant speaks one language; Salesforce Marketing Cloud speaks another. MCP sits between them and handles the translation — in real time, with context intact.

MCP vs. Traditional API Integrations

Most teams have already built API integrations into SFMC. Understanding how MCP is different is key to scoping your first use cases. 

A traditional API call says: “Do exactly this.” 

An MCP-enabled agent says: “Here’s my goal. Figure out the steps.” 

Why MOps Teams Are Paying Attention

Over the past two years, AI in marketing moved from assistants to agents, i.e., from tools that help you think to tools that act. That shift has quietly raised the bar for what a martech platform needs to support.

It’s also changed how marketers expect to work. Conversational workflows are no longer novel. Describing a goal and having it executed is the default in most AI-native surfaces. 

As platforms adapt to those expectations, the Marketing Cloud MCP Server is  Salesforce’s first concrete deliverable in that direction. And the timing is deliberate: it meets teams exactly where their workflows have arrived, be it Slack, Claude, or a CLI. 

What Is the Salesforce Marketing Cloud MCP Server?

Architecture and Core Concept

The MCP Server for Marketing Cloud Engagement is a Salesforce-hosted bridge between your AI assistant and your SFMC instance. Three components make it work:

• The LLM (Claude, Gemini, etc.): Interprets your request and decides which tools to call
• The MCP Client (Claude Code, Gemini CLI, Claude.ai): The interface you type into
• The MCP Server: Receives requests from the LLM, executes functions in Marketing Cloud, and returns results

The server is not a storage layer. It doesn’t hold your data between sessions. It’s a temporary, authenticated interface that opens when a user initiates it and closes when the session ends. 

What It Can Do

The MCP Server covers six areas of Marketing Cloud operations: 

Journey Builder Management: Create, update, pause, resume, and stop journeys without opening the UI. Insert or exit contacts from active journeys, individually or in bulk. Build full-journey activity payloads that include email steps, wait activities, decision splits, and random splits.

Email and Messaging: Send transactional emails via triggered send definitions. Create and manage email assets in Content Builder. Build templates and HTML email assets. Refresh content in running journeys without republishing.

SMS and Mobile: Create transactional SMS definitions and send outbound SMS immediately. Manage MobileConnect keywords, retrieve short/long code configurations, and check SMS subscription status.

Data Extensions and Contacts: Create, update, and delete data extensions, including field management. Upsert and retrieve individual records. Update contact attributes across profile and linked data extensions. Query opt-in status by email address.

Automations and SQL: Create, run, update, and retrieve automations and their execution instances. Write and run SQL Query activities with built-in validation before execution. It is particularly useful for admins who spend time debugging complex queries.

Analytics and Content Builder: Retrieve transactional send tracking within date ranges. Get triggered send summary metrics. Search, create, and update content assets. Manage folders and event definitions for API-triggered journeys.

Exploring the Practical Applications of the SFMC MCP Server with Claude 

The Grazitti team connected a live Marketing Cloud instance to Claude via the MCP Server and put it through real-world scenarios: 

Conversational Audience Segmentation

A MOps analyst needs a sendable data extension for a new loyalty campaign. 

Traditional way: Navigate to Contact Builder, define fields, set data types, configure the send relationship. 

Via MCP: Type a plain English description. The agent maps field descriptions to Marketing Cloud data types, sets the primary key, and returns a direct link to the new data extension.

We tested this using the following prompt:

Create a data extension named “MCP_Server_Test” in folder “Grazitti_Test” with fields details as  [“SubKey”,”Text”,18],[“Email”,”EmailAddress”,254],[“Name”,”Text”,50] . Configuring SubKey field as Primarykey and also as Send relationship field to All Subscribers List. 

The data extension was created correctly, instantly, with all field configurations intact.

AI-Assisted Journey Orchestration

We then built a complete engagement journey through a single prompt, without any Journey Builder navigation:

Build a journey named ‘Claude_Test_Journey’ firing on MCP_Server_Test, running daily at 12 PM IST with new records only. Send ‘Grazitti_Test’ email, wait 3 days, split on open behavior. Send ‘WelcomeEmail_POC’ to openers, wait 1 day, exit. Re-send ‘Grazitti_Test’ to non-openers, wait 2 days, exit. Set re-entry to ‘Re-entry only after exit.'”

The journey was built successfully with entry source, send cadence, decision split logic, and re-entry rules, all configured.

Work that would take a mid-level SFMC admin 45 minutes to an hour was done in one session.

SFMC MCP Server vs. Agentforce vs. Einstein AI

This is the question almost every MOps leader asks. The confusion is real enough that Martin Kihn, SVP Strategy at Salesforce, felt compelled to address it directly on LinkedIn.

Here are the key differences among the three approaches of AI in SFMC.

Einstein AI makes your people smarter. Agentforce removes people from certain task categories entirely. The MCP Server accelerates the work your people are already doing.

When to Use Each

Use Einstein when you need predictive intelligence embedded in your campaigns. Example: send time optimization, engagement scoring, and content recommendations.

Use Agentforce when you need autonomous, customer-facing orchestration at scale. Example: two-way SMS conversations, real-time journey decisions, live agent handoffs.

Use the MCP Server when your team needs to build, configure, and operate Marketing Cloud faster. Example: data extension setup, journey creation, campaign QA, SQL generation, reporting.

Can They Work Together?

Yes, and this is the right framing. Salesforce’s own description of the relationship: MCP helps teams build more campaigns, more journeys, more personalized content. Agentforce activates that content in customer interactions. The output of one feeds the other. More built means more to orchestrate, which is the flywheel Salesforce is designing for.

How to Set Up the SFMC MCP Server

The setup follows a standard installed package flow. If your team has configured API integrations in Marketing Cloud before, this will be familiar ground.

Prerequisites: Install and configure an MCP-compatible AI client (Claude Code or Gemini CLI) before touching Marketing Cloud setup.

Step 1 — Create an Installed Package: In SFMC Setup, search “Installed Packages” in the Quick Find box. 

Create a new package, give it a descriptive name (e.g., MCP_SFMC_Connector), and save.

Step 2 — Add an API Integration Component: On the package detail page, click Add Component. 

Add an API Integration component by choosing API Integration. 

Then choose Public App as the integration type.  Enter https://salesforce.com as a placeholder redirect URI for now — you’ll replace this later.

Step 3 — Assign Permission Scopes: Choose your permission scopes carefully and allow only the ones which are required. The scopes you grant here define what the AI assistant can do in your account.

Step 4 — Retrieve Client ID and Tenant ID: From the package detail page, locate your Client ID (24-character alphanumeric string) and Tenant ID (the 28-character subdomain from your authentication URI). For example, if your Authentication Base URI is https://mcphchq9d5u2uk2egt56v.auth.marketingcloudapis.com/, your Tenant ID is mcphchq9d5u2uk2egt56v.

Step 5 — Update the Redirect URI: Edit the API Integration and replace the placeholder with the regional callback URL:

• US: https://mai-mce-mcp-cdp1.sfdc-yfeipo.svc.sfdcfc.net/t/{tenantId}/c/{clientId}/api/mcp/oauth/callback
• EU: https://mai-mce-mcp-cdp1.sfdc-yzvdd4.svc.sfdcfc.net/t/{tenantId}/c/{clientId}/api/mcp/oauth/callback

Substitute your actual tenant and client IDs. Select the endpoint nearest to your region to minimize latency.

Step 6 — Connect Your AI Client: In Claude (or Claude Code), go to Customize → Connectors → Add Connector. Enter the MCP server URL:

https://mai-mce-mcp-cdp1.sfdc-yfeipo.svc.sfdcfc.net/t/{tenantID}/c/{clientID}/api/mcp

The client will prompt you to authenticate against your SFMC account using admin credentials.

Step 7 — Set Tool Permission Levels: Once connected, set permission levels for read-only and write/delete tools: Always Allow, Needs Approval, Blocked, or Custom. For initial deployments, set write/delete tools to Needs Approval until your team is comfortable with the agent’s behavior.

AI assistants can produce inaccurate or harmful results. Assign the assistant only the permission scopes necessary for it to complete the tasks you ask it to. Review its responses to your requests carefully for accuracy and safety. You assume responsibility for how the outcomes of AI are applied to your organization. 

Sanity Check Prompts: Before using in production, validate with low-stakes tests:

• “Create a Data Extension called ‘Test_DE’ with fields for email (EmailAddress), key (Text, 18), and name (Text, 50).”

• “Retrieve subscriber status for [email address].”

Where Does Salesforce Stand on MCP Today?

Native MCP: Already GA

The MCP Server for Marketing Cloud Engagement is generally available as of Summer ’26. It is Salesforce-hosted, with no additional license cost beyond LLM token consumption. It ships with US and EU regional endpoints for latency optimization. 

Headless 360 and the Bigger Bet

It sits within Headless 360, Salesforce’s architectural shift that restructures the platform into four independently accessible layers: Data 360, Business Logic, Agentforce Orchestration, and Engagement. The MCP Server lives in the Engagement layer.  

Community and Third-Party Servers

Beyond the native server, open-source MCP implementations for SFMC exist in the developer community, and third-party providers offer their own connectors. For most enterprise deployments, the native server is the right starting point. It inherits the platform’s existing permission model, runs on Salesforce infrastructure, and requires no additional vendor relationships. 

Current Limitations

The category is still maturing. The server wraps existing Marketing Cloud Engagement API functionality and doesn’t yet cover everything. Salesforce is actively soliciting feedback on missing tools via IdeaExchange. Teams should expect the tool library to expand and should build their use case roadmap accordingly. 

The Need for Human Oversight 

AI assistants and large language models (LLMs) significantly improve efficiency and reduce manual effort. However, it is essential to exercise caution by validating both the prompts provided and the outputs generated by the AI.  This is to ensure that inaccurate communication or data is not inadvertently sent to customers. All AI-driven configurations should continue to involve human review and oversight. 

Is the SFMC MCP Server Right for Your Team?

When It Makes Sense

Teams that get value from this quickly are the ones who’ve already built solid data extension hygiene. This includes clean schemas, consistent naming conventions, and documented journeys. They have AI initiatives in motion and need SFMC to plug into those workflows. 

When It May Be Too Early

Poor data quality is the most common derailment. If your data house isn’t in order, the MCP Server amplifies the problem.

Small teams with minimal automation maturity may find the setup-to-value ratio unfavorable at this stage. Similarly, teams that have recently invested in native AI capabilities like Agentforce or Einstein might want to stabilize those deployments before adding another AI layer to their stack.

The Bottom Line

The Marketing Cloud MCP Server changes the unit of effort for MOps work. Tasks that required UI navigation, developer involvement, or multiple tools now resolve in a single prompt. That’s not a small shift. It compounds across every campaign, every journey, every data request your team handles.

The ecosystem is still maturing. The teams that build competency now will operate differently from those that wait.

The window is still open — but not indefinitely.

Looking To Connect Your SFMC Instance to an AI Workflow? Talk to Our Marketing Cloud Team.

We’ll help you assess readiness, scope the right use cases, and implement the SFMC MCP Server with the governance foundation your data deserves. For more information, email us at [email protected].