Salesforce Expands Headless 360 With MCP for Agentic CX

What happened

Salesforce has expanded its Headless 360 offering with an architecture built on the Model Context Protocol (MCP), enabling AI agents to discover and act on capabilities across enterprise workflows at runtime. The update moves headless CRM beyond simply exposing Salesforce functionality outside its standard interface, giving agents governed access to customer data, business logic and executable actions rather than static integration points.

According to CX Today, the change effectively turns previously siloed Salesforce capabilities into resources that autonomous agents can query and act on dynamically, rather than requiring developers to hard-code each integration in advance.

Why it matters

This is fundamentally a technology-architecture story: MCP support signals Salesforce's move to make its platform "agent-ready" at a structural level, not just through point features. By exposing data, logic and permissioned actions through a discoverable protocol, the update changes how agentic systems are built — shifting integration work from bespoke coding toward standardised, governed access that agents can navigate themselves.

For enterprise and CX leaders, the practical question becomes whether existing workflows, permissions and data models are structured well enough for an agent to use them safely and sensibly at runtime. Agentic AI is only as good as the scaffolding beneath it — and this expansion pushes that scaffolding question higher up the leadership agenda.

The Renascence take

Most coverage of agentic AI focuses on what agents can do once unleashed. The more consequential question is what they are unleashed onto — and this update puts that infrastructure layer in plain sight.

Agentic CX will not be won or lost on model quality; it will be won or lost on whether the underlying business logic, data governance and permissioning were designed for autonomous use in the first place. Many organisations have spent years building workflows for human operators — full of implicit judgement calls, exceptions and tribal knowledge — that simply do not translate cleanly into something an agent can execute safely. Before adopting agent-ready architectures like this one, customer-obsessed operators should audit their existing processes for hidden human-only logic, not just plug in the new protocol and assume the agent will figure it out.