What AIforce Means for Marketing Cloud and Account Engagement (Pardot) Teams

For marketing teams, AIforce is mostly about what is coming next, with one important exception.

Salesforce in Claude, the first Claudeforce product, launched with sales skills only. Salesforce says skills for marketing, service, commerce and industries will follow “in the near future,” without a date. But marketers do not have to wait:

• Marketing Cloud Engagement already has a generally available MCP server that lets Claude and other AI agents manage data extensions, journeys and automations in plain language.
• Marketing Cloud Next gets “Headless Marketing,” MCP access to journeys, flows and messaging, with general availability planned for October 2026.
• Account Engagement (Pardot) has new Agentforce actions but no official MCP server yet.

This guide explains what AIforce is from a marketer’s point of view, what is available today for each Marketing Cloud product, and what is coming. It also covers how to prepare without taking risks with your sends, your data or your consent records.

Key takeaways

• Salesforce in Claude has no marketing skills yet. Salesforce says they are coming, with no date.
• Marketing Cloud Engagement teams can already connect Claude and other AI agents through Salesforce’s first-party MCE MCP Server, generally available since June 2026.
• Marketing Cloud Next is getting Headless Marketing, Campaign Agent and an Account Discovery Agent, all scheduled for general availability in October 2026.
• Account Engagement has six Agentforce actions for prospects, lists and Engagement Studio, but no official MCP server. Treat community MCP servers as third-party software that needs a security review.
• The biggest risk for marketing is an agent triggering a send or changing data at scale. Scope permissions narrowly, and require confirmation for anything that sends, deletes or changes consent.

AIforce in 90 seconds, for marketers

AIforce is the interface layer Salesforce announced at Dreamforce on September 15, 2026. It brings Salesforce data, workflows, business logic and permissions into AI tools such as Claude, Slack and Salesforce’s own Lightning interface. People can ask questions, update records and trigger workflows without navigating Salesforce’s screens. It launched with three interfaces:

• Claudeforce: the Salesforce and Anthropic partnership. Its first product, Salesforce in Claude, is a plugin with 37 prebuilt sales skills.
• Slackforce: Salesforce inside Slack, including Slackbot, live interactive “Surfaces,” Slack CRM and Slack Code.
• Agentforce Coworker: an AI teammate in Lightning.

AIforce sits on what Salesforce calls the Headless Toolkit: MCP servers, APIs, plugins and skills that expose the platform to AI agents. For marketing, that toolkit matters more today than the three branded interfaces, because it is how Marketing Cloud is already becoming accessible from Claude and other AI tools.

A note on MCP: the Model Context Protocol is an open standard that lets AI agents connect to external systems through defined “tools.” Salesforce’s Marketing Cloud engineers describe it as a universal adapter for AI. When a Salesforce product has an MCP server, an AI agent such as Claude can read and act on that product in plain language, within the permissions you grant.

First, the naming: which Marketing Cloud are we talking about?

Salesforce spent 2025 and 2026 renaming its marketing products, then partly reversed course after Dreamforce 2026. Here is the current map.

The distinction matters because AI capabilities differ sharply by product. Engagement has a GA MCP server, Next is getting agents and headless access in October, and Account Engagement is furthest behind on MCP.

What is live today and what is coming

The table below combines Salesforce’s Marketing Cloud announcements from Dreamforce 2026 with earlier 2026 releases.

GA dates reflect Salesforce’s plans as published in September 2026 and may move.

For Marketing Cloud Engagement teams: use the MCE MCP Server now

Salesforce released the MCE MCP Server to general availability on June 2, 2026. It describes the server as its first-party, hosted MCP server for Marketing Cloud Engagement, and a first step toward a headless experience. It exposes Engagement’s core capabilities as tools that any MCP-compatible agent, including Claude, can call without writing code or opening the Marketing Cloud UI.

How permissions work

Understand this part before anything else. An agent’s access is the overlap of two things:

• the scopes on a dedicated installed package in your Marketing Cloud account, and
• the permissions of the user who signs in.

Salesforce’s example: if the installed package can read data extensions but the user cannot, the agent cannot either. The reverse also holds, so you can restrict an agent simply by not granting scopes to the installed package, even for users who have broader rights.

That gives you two controls. Use the installed package to set the ceiling for what any agent can do, and user permissions to limit individuals below that ceiling.

How setup works

1. Create a dedicated installed package for AI agent access and grant only the scopes you intend agents to use.
2. Get the MCP URL from the installed package. Salesforce notes it is specific to your organization and package, and not sensitive in itself.
3. Register the URL in your AI client. In Claude Code this is a single command; other clients add it as a connector.
4. Sign in to Marketing Cloud from inside the client when prompted. The agent receives a time-limited token to act on your behalf.

Connector steps differ by AI client, so follow Salesforce’s MCE MCP setup guidance for the client you use. Marketing Cloud API limits and guidelines still apply to everything the agent does.

What marketers can do with it

The server wraps existing Marketing Cloud Engagement APIs, so the agent can do in conversation what your team currently does in Contact Builder, Journey Builder and Automation Studio. Salesforce’s engineering team tested examples like these:

• Create a data extension from a description. Describe the fields you need, and the agent maps them to Marketing Cloud data types, sets the primary key and returns a link to the new data extension.
• Build a simple welcome journey. The agent checks the entry source, builds the entry, email and exit steps, attaches the email asset and saves the journey as a draft, then asks whether to activate it.
• Build a multi-email nurture series. Describe the sequence and waits in plain language, and the agent builds each step and returns the Journey Builder link for review.
• Add Einstein Send Time Optimization before a specific email, after checking that the feature is provisioned for your business unit.
• Propagate a new field across dependent data extensions and queries. The agent traces dependencies, shows an impact report and makes the changes only after you confirm, with a dry-run option to preview everything first.

The last example is where agents earn their keep. Tracing a field through every downstream data extension and SQL query activity is tedious and easy to get wrong by hand.

The warning Salesforce gives, and we agree with

Salesforce chose to support destructive operations such as deleting data. They are labeled as destructive in a way the agent can read, but the agent can still run them if it has permission. Salesforce’s engineers advise thinking through the worst case when assigning permissions, and point out that an unintended send can cause reputational harm. For marketing, that is the risk that matters most: an agent that can send to a production audience can send to the wrong one.

Our recommendation for a first pilot:

• Start in a non-production business unit, or with read-only scopes in production.
• Do not grant send, delete or journey-activation scopes to the pilot’s installed package.
• Keep journeys the agent creates in draft, and activate them manually after review.
• Log what the agent does and review it weekly.

For Marketing Cloud Next teams: Headless Marketing and agents in October

Marketing Cloud Next is where Salesforce is putting most of its new marketing AI. Its Dreamforce 2026 announcements include:

• Headless Marketing (GA planned October 2026): Salesforce says a new MCP-powered architecture will let you and any AI model work with Marketing Cloud in plain language, across journeys, flows and messaging, from any surface. This is the marketing equivalent of what Headless 360 did for Salesforce’s CRM, and it is the most likely path for Claude to work with Marketing Cloud Next.
• Campaign Agent (GA planned October 2026): Salesforce describes it as a digital campaign manager that learns your goals, guardrails and brand guidelines, and can create, run and optimize campaigns. It generates and tailors content by channel, assembles audiences, content, journeys and channels toward a goal, and selects which variation and journey each customer receives.
• Account Discovery Agent (GA planned October 2026): finds high-potential accounts and buying groups for B2B outreach.
• Personalized Paths in Flow and Mobile Flash Sending (GA planned October 2026), and RCS Messaging (GA now).

Marketing Cloud Next runs on Data 360, which means the quality of your unified profiles decides how well these agents perform. Data company Acxiom made the same point in its Dreamforce 2026 takeaways: if the underlying customer profile is incomplete or fragmented, automation makes the wrong decision faster and at greater scale. Before October, audit identity resolution rules, consent data and the fields your journeys depend on. If your unified profiles need work, a Salesforce Data Cloud (Data 360) implementation review is the right starting point.

For Account Engagement (Pardot) teams: the honest picture

Account Engagement is the product where marketers will find the least AIforce capability today. Here is what exists.

Agentforce actions for Account Engagement

Salesforce’s Summer ’26 release added six Account Engagement actions for Agentforce. Salesforce’s help documentation on using Account Engagement actions in Agentforce says they let you create a dedicated Account Engagement Lead Management Agent. That agent can view and update prospect records and add prospects to nurture campaigns from anywhere in Salesforce. The underlying actions cover:

• Viewing a prospect’s details.
• Creating or updating a prospect.
• Listing marketing lists.
• Checking a prospect’s list memberships.
• Signing a prospect up to marketing lists.
• Retrieving Engagement Studio programs.

Requirements, per Salesforce:

• Account Engagement with Salesforce Enterprise, Performance or Unlimited edition.
• The latest Account Engagement (Pardot) AppExchange package.
• Agentforce enabled.
• An External Client App plus a Named Credential.
• Access to the underlying Apex classes for users who deploy the actions.

These actions run inside Agentforce. They are not part of Salesforce in Claude, and they do not make Account Engagement available as an MCP server for Claude. For a related use case, see our guide to Agentforce and Account Engagement lead scoring.

No official MCP server yet

As of September 2026, Salesforce has not published an official MCP server for Account Engagement. Community-built servers exist, including open-source projects that wrap the Account Engagement v5 API and commercial connectors. Treat them as third-party software:

• Review who maintains the code and how often it is updated.
• Check how credentials are stored. Some use OAuth client credentials that act as a service account rather than as each user, which bypasses the per-user permission model Salesforce uses for its hosted MCP servers.
• Prefer read-only access. The Account Engagement API has daily request limits that vary by edition, and an agent can consume them quickly.
• Get your security team’s approval before connecting any of them to production.

Moving toward Marketing Cloud Next

Salesforce’s certification material for Summer ’26 highlights Marketing Cloud Next for Account Engagement, which extends approved content delivery across email, SMS and WhatsApp. Salesforce Ben noted that Summer ’26 also simplified consent management between marketing products, which it described as one of the biggest barriers to moving to Marketing Cloud Next. If your B2B marketing roadmap depends on the October agents and Headless Marketing, plan the path from Account Engagement toward Marketing Cloud Next now. Our Marketing Cloud Account Engagement services team can help map that transition.

Sales skills marketers can use today

Salesforce in Claude has no marketing skills, but several of its sales skills overlap with marketing work. Marketers who have a Sales Cloud license and access to the plugin can use them today. From Salesforce’s public plugin repository:

All of these run under the user’s own Salesforce permissions and do not touch Marketing Cloud.

Governance rules for AI agents in marketing

Sales agents mostly read and update records. Marketing agents can send messages to thousands of people. That changes the controls you need.

• Separate “build” from “send.” Let agents create data extensions, draft journeys and assemble content. Keep activation and sending as a human step until you have months of evidence.
• Scope by business unit. Pilot in a non-production or low-risk business unit first.
• Protect consent. Never grant an agent the ability to change subscription status or consent fields without review. A wrong consent change is a compliance problem, not just a data problem.
• Use dry runs. Salesforce’s MCE MCP Server supports previewing bulk changes before applying them. Make it standard practice.
• Give agents your brand rules. Campaign Agent is designed to take brand guidelines as input. Write them down clearly now, including banned phrases, tone and legal disclaimers.
• Watch API consumption. Marketing Cloud and Account Engagement API limits apply to agent activity.
• Keep an audit trail. Know which user’s token did what, and review it.

A 90-day plan for marketing operations

Days 1 to 30: Foundations

• Map which Marketing Cloud products you run and where each sits in the table above.
• Audit consent data, identity resolution and the fields your key journeys depend on.
• Write brand, tone and compliance guidelines in a format an agent can use.
• If you run Engagement, create a pilot installed package with read-only scopes.

Days 31 to 60: Controlled pilots

• Engagement: connect Claude to the MCE MCP Server in a non-production business unit. Test data extension creation, draft journeys and dependency tracing.
• Account Engagement: evaluate the Agentforce actions with a Lead Management Agent for a small sales-and-marketing team.
• Sales-marketing overlap: if you have access to Salesforce in Claude, run customer-voice and win-loss reviews through it to feed content planning.

Days 61 to 90: Prepare for October releases

• Marketing Cloud Next teams: plan Campaign Agent and Headless Marketing pilots for their GA dates.
• Account Engagement teams: build the business case and timeline for Marketing Cloud Next.
• Everyone: watch for Salesforce’s marketing skills for Salesforce in Claude, and test them on day one.

What to watch next

• Marketing skills for Salesforce in Claude. Salesforce has said they are coming. When they ship, expect prebuilt workflows similar to the sales skills.
• MCE MCP Server coverage. Salesforce says it plans to extend the server to more Marketing Cloud APIs over time.
• Headless Marketing GA. Scheduled for October 2026.
• An official Account Engagement MCP server. Not announced as of this writing.
• AI search visibility tooling. Salesforce’s marketing announcements now promote Palmata, which reports how a brand appears in AI answer engines and suggests content changes. Search visibility in AI tools is becoming part of the marketing stack, not a separate SEO concern.

Get your marketing stack ready

HyphenX Solutions implements and optimizes Marketing Cloud and Account Engagement, and helps teams pilot AI agents safely, from installed package scoping to consent audits and Marketing Cloud Next planning. Talk to our Salesforce Marketing Cloud services team to get your marketing stack ready for AIforce.

FAQs

Does Salesforce in Claude work with Marketing Cloud?

Not yet. Salesforce in Claude, the first Claudeforce product, launched in beta on September 15, 2026 with sales skills only. Salesforce says skills for marketing, service, commerce and industries will follow in the near future, but has not given a date. Marketing Cloud Engagement teams do not have to wait, though. Salesforce’s separate MCE MCP Server, generally available since June 2026, already lets Claude and other MCP-compatible AI agents work with data extensions, journeys and automations in plain language, within the permissions you grant.

Is there an MCP server for Marketing Cloud?

Yes, for Marketing Cloud Engagement. Salesforce released its first-party, hosted MCE MCP Server to general availability on June 2, 2026. It exposes Engagement’s core capabilities, including data extensions, journeys and automations, as tools that Claude and other MCP-compatible agents can call. Salesforce says it plans to extend the server to more APIs over time. For Marketing Cloud Next, Salesforce has announced Headless Marketing, an MCP-powered architecture for journeys, flows and messaging, with general availability planned for October 2026. Account Engagement does not yet have an official MCP server.

Is there an MCP server for Account Engagement (Pardot)?

Not an official one as of September 2026. Salesforce has not published a first-party MCP server for Account Engagement. Community-built servers exist, including open-source projects that wrap the Account Engagement v5 API and commercial connectors. Treat them as third-party software: review who maintains them, and prefer read-only access. Check how credentials are stored, since some act as a service account rather than as each user. Account Engagement API daily limits also apply. Get your security team’s approval before connecting any of them to production data.

What can Agentforce do in Account Engagement?

Salesforce’s Summer ’26 release added six Account Engagement actions for Agentforce. They let you build a dedicated Lead Management Agent that works from anywhere in Salesforce. It can view a prospect’s details, create or update prospects, list marketing lists, check a prospect’s list memberships, sign prospects up to lists and retrieve Engagement Studio programs. Requirements include Enterprise, Performance or Unlimited edition, the latest Account Engagement package, Agentforce enabled, and an External Client App with a Named Credential. These actions run inside Agentforce, not in Claude.

Can an AI agent send emails from Marketing Cloud?

If you grant it the permissions, yes. Salesforce’s MCE MCP Server supports a wide range of operations, including destructive ones such as deleting data. These are labeled as destructive so the agent can recognize them, but it can still run them if it has permission. Salesforce’s own engineers warn that an unintended send can cause reputational harm. During pilots, keep sends and journey activation as human steps, leave send and delete scopes off the installed package, and keep agent-built journeys in draft until reviewed.

Is Agentforce Marketing the same as Marketing Cloud?

Yes. Agentforce Marketing was the umbrella name Salesforce used for its marketing product family after Dreamforce 2025, and around Dreamforce 2026 that umbrella returned to being called Marketing Cloud. Individual product names did not change: Marketing Cloud Engagement, Marketing Cloud Account Engagement (formerly Pardot) and Marketing Cloud Next all keep their names. The AI capabilities announced under the Agentforce Marketing name continue, and the Agentforce agent platform itself also kept its name. Documents using either name refer to the same products.

What is Headless Marketing in Marketing Cloud Next?

Headless Marketing is a new MCP-powered architecture for Marketing Cloud Next, with general availability planned for October 2026. Salesforce says it will let marketers and any AI model work with Marketing Cloud in plain language across journeys, flows and messaging, from any surface. It is the marketing equivalent of what Headless 360 did for Salesforce’s CRM. Once it ships, it is the most likely route for Claude and other AI tools to work directly with Marketing Cloud Next. It depends on clean Data 360 profiles to perform well.

What is Campaign Agent in Marketing Cloud Next?

Campaign Agent is an AI agent for Marketing Cloud Next, with general availability planned for October 2026. Salesforce describes it as a digital campaign manager that learns your goals, guardrails and brand guidelines, then creates, runs and optimizes campaigns. It generates and tailors content by channel, assembles audiences, content, journeys and channels toward a goal, and selects which variation and journey each customer receives. To get value from it, write clear brand rules now and audit the Data 360 profiles it will rely on.

How do permissions work for AI agents in Marketing Cloud Engagement?

An agent’s access is the overlap of two things: the scopes on a dedicated installed package in your Marketing Cloud account, and the permissions of the user who signs in. If the package allows reading data extensions but the user cannot read them, the agent cannot either, and the reverse also holds. Use the installed package to set the ceiling for what any agent can do, and user permissions to limit individuals below it. For pilots, grant read-only or build-only scopes, and leave send and delete scopes out entirely.

How should marketing teams prepare for AIforce?

Start with foundations. Map which Marketing Cloud products you run, since AI capabilities differ sharply between Engagement, Next and Account Engagement. Audit consent data, identity resolution and the fields your key journeys depend on. Write brand, tone and compliance guidelines in a form an agent can use. Then run controlled pilots: the MCE MCP Server in a non-production business unit, or Account Engagement’s Agentforce actions with a small team. Plan Campaign Agent and Headless Marketing pilots around their October 2026 release dates.