Understanding Headless 360 MCP Server (Beta Version): What is it & Why it Matters

In July 2026,Salesforce launched its most recent update “Headless 360 MCP Server”. This simply means that the developers, admins, and other linked parties get a unified interface to integrate and configure the activities handled by AI bots.

The Headless 360 MCP Server operates on the previously built Model Context Protocol (MCP), which is an open protocol allowing AI tools to interact with other third-party services via a uniform interface, and is an addition to the Hosted MCP Servers that Salesforce launched in April 2026.

This blog aims to provide an overview of what this server does, its tool architecture, and control mechanisms.

What is MCP, in simple terms?

What is MCP? It is an abbreviation for the term Model Context Protocol. Imagine it as a universal means by which AI agents (e.g., Claude, any other AI assistant) can communicate with a software system and perform various actions within it, not just give information about it.

In the absence of MCP connections, an AI assistant could provide you with an explanation of how to make some action on Salesforce, but it couldn’t make the action on your behalf. Now, with MCP connections, an AI agent is able to log into Salesforce (if granted access), view the available tools, and perform such actions as creating a user, writing some code, or creating an integration.

The Headless 360 MCP Server is Salesforce’s solution of providing such MCP connection with practically all features of Salesforce through one entry point.

Why “Headless 360”?

Headless means there is no screen or user interface involved. The agent works in the background, not by clicking through Salesforce’s website.
360 signals that it covers the full range of Salesforce, not just one product or feature.

So instead of an AI agent clicking around Salesforce Setup screens the way a human would, it can now reach into Salesforce directly and complete tasks.

The core problem Salesforce was solving

Salesforce boasts of thousands of features. Had Salesforce made all those features into individual tools that can be used by AI agents, the AI agents would have been required to go through thousands of descriptions in order to determine which tool was to be used for an assigned task.

That is slow, expensive (because AI usage is billed partly on how much text it has to process), and it gets worse every time Salesforce adds new features.

The fix: four tools instead of thousands

Rather than exposing every feature separately, Salesforce built the Headless 360 Server around four tools. Behind those four tools sits a large and growing library of “skills,” which are like instruction packets the AI can pull up when needed.

Tool	What it does	Think of it like
Discover	Searches a library of Salesforce skills to find ones that match what you asked for	A smart search engine for “what can I use to do this?”
Describe	Pulls up the full technical details of a chosen skill, including what steps need to happen and in what order	An instruction manual for one specific task
Dispatch	Actually carries out the skill, calling the right systems in the right order	The “go” button
Dispatch Read Only	Same as Dispatch, but only for tasks that just read or look at data, without changing anything	A safe, look-only version of Dispatch

A real example of how this works

Let’s assume that you instruct your AI agent to “plan an integration that publishes events to AWS when Account records get updated.” In this case, normally, a user will have to manually navigate through three different Setup pages to achieve the goal.

These are the actions performed by your agent:

The agent starts with the Discover operation to look for related skills: change data capture, named credentials, and event relay.
The agent then moves to the Describe operation to determine the list of API operations to invoke and their order. For example, the named credential needs to be created before the event relay can be set up.
The agent finally uses Dispatch operation to perform the actual invocation of these API operations.

The agent essentially does the multi-step clicking and configuring that a person would otherwise do by hand.

What can it actually do today?

Salesforce is launching the Beta with around 100 skills, with thousands more planned. At launch, most of these skills focus on admin-style Setup tasks. A few of the most common ones include:

Managing users: creating, deactivating, freezing or unfreezing accounts, resetting passwords, and assigning permission sets or licenses
Managing Apex triggers: reading, writing, and deploying Apex code triggers
Building event-driven integrations: setting up platform events, Change Data Capture, and event relays
Creating named credentials: setting authentication methods, endpoints, and certificates

Over time during the Beta, Salesforce plans to expand coverage beyond admin tasks to skills that regular business users rely on across different Salesforce products.

Where do all these skills come from?

Salesforce developed an internal framework known as the “Headless 360 Factory”. This framework analyzes Salesforce Setup for any feature which does not yet have API access. Subsequently, it automates generation of the APIs and skills required by an AI agent to leverage these features. The skills generated by AI are then reviewed by Salesforce product teams and skills added by them, giving a complete platform representation in one place.

Is this safe? What stops an AI agent from doing something it should not?

This is probably the most important part for anyone responsible for a Salesforce org. Salesforce built in several layers of protection.

Safeguard	What it means
Runs as the authenticated user	The AI agent can only do what the actual logged-in person is allowed to do. Nothing more.
Standard permission rules still apply	Regular Salesforce permissions such as CRUD access, field-level security, sharing rules, profiles, and permission sets are all enforced exactly as they would be for a human user.
Full audit trail	Every action the agent takes is logged and tied back to the person who authorized it.
Tool-level approval settings	Most AI clients let you require the agent to ask for your approval before it runs a tool, which is especially useful for admins and developers with broad access.

In short, if a person could not normally do something inside Salesforce, their AI agent cannot do it either.

Salesforce also recommends the same best practice that applies to any Salesforce configuration change: test in a sandbox or developer org first, then move to production once you are confident.

How is this different from what came before?

This builds on something Salesforce already launched earlier in the year: Hosted MCP Servers, which became generally available in April 2026. The Headless 360 MCP Server is described as the “next big step” on that same path, expanding what those hosted servers can reach and do.

How to try it

Getting access to the Beta is straightforward:

Go to Salesforce Setup.
Look for the Salesforce Hosted MCP Servers section.
Find Headless 360 in the list.
Click Activate.

Salesforce is also asking for feedback during the Beta and has set up a GitHub repository where users can post questions, report issues, and share comments.

What this means depending on your role

Not everyone will use this feature the same way. Here is a quick breakdown of what changes for different people working with Salesforce.

Role	What changes for you
Salesforce Admin	Activities that require repetitive setup, such as user creation and named credential configuration, could be automated via AI agents rather than manual actions on each individual screen.
Developer	Apex triggers can be read, written, and deployed through an agent, which can speed up repetitive coding and deployment work.
Business user	Not very important for today’s launches since most of the skills available are related to the admin process, but Salesforce plans to cover normal day-to-day business operations in their Beta program.
IT and security teams	Every action is still tied to a real user’s permissions and shows up in the audit trail, so oversight does not disappear just because an AI is involved.

Conclusion – The Bigger Picture

The Headless 360 MCP Server announcement is part of a wider shift happening across the software industry, where AI agents are moving from simply answering questions to actually completing multi-step tasks inside real business systems.

Salesforce’s approach of using a small, stable set of tools backed by a large, growing skill library is one way to make that shift scalable, instead of forcing AI models to wade through an ever-expanding pile of separate tools.

For Salesforce administrators, developers, and other business personnel, the key point is that tasks which used to require many manual steps on different interfaces can now potentially be done by the AI agent, all with the same permissions and security that apply to regular human users.

The Beta is still early, with about 100 skills live today and thousands more planned. It is a space worth watching if your team relies on Salesforce for day-to-day operations.

FAQs

Is this going to allow an AI agent to do something that a human cannot do in Salesforce?

No. All activities are executed on behalf of the authenticated user, so all permission settings such as field-level security, sharing rules, and profiles continue to govern access.

Am I required to write some code to make use of it?

No. Once the Beta version is enabled in Setup, any MCP-compatible AI clients will be able to communicate with it and make use of all four tools. Building your own custom skills is an optional step intended for developers.

Is this feature relevant only to large and complicated Salesforce organizations?

No, it may be valuable even for smaller organizations that want to automate repetitive configuration activities, though the most obvious benefits will be available for organizations with a lot of configurations or integrations.

Are new skills going to be released in the future?

Yes. Salesforce starts with a few hundred skills but aims to have thousands of new skills.

For more insights, updates, and expert tips, follow us on LinkedIn.