Salesforce Headless 360 Adds MCP for Agentic CX

What happened

Salesforce has expanded its Headless 360 offering with a new architecture built on the Model Context Protocol (MCP), giving AI agents the ability to discover and act on existing enterprise capabilities across customer workflows. The update, reported by CX Today alongside a roundup of other developments from Stripe, OpenAI and Anthropic, positions Headless 360 as infrastructure for what Salesforce is calling "agentic CX" — service and sales workflows that AI agents can navigate and execute against without needing bespoke integration work for every system they touch.

The MCP-based approach allows agents to identify what functions and data are available within an enterprise's existing tech stack, then act on that inventory directly, rather than requiring each capability to be manually wired into an agent's toolset. Details on the accompanying Stripe, OpenAI and Anthropic developments referenced in the same roundup were not elaborated in the available reporting.

Why it matters

MCP has quickly become a de facto standard for letting AI agents talk to enterprise systems, and Salesforce folding it into Headless 360 signals that the industry is moving from single-purpose chatbots toward agents that can operate across a company's full stack of customer-facing tools. For organisations running composable or headless CX architectures, this reduces one of the biggest blockers to agentic deployment: the integration overhead of connecting an agent to every backend system it might need to use.

The practical effect is that enterprises with modular commerce, service or CRM architectures are better positioned to let agents act autonomously — checking order status, updating records, triggering workflows — using capabilities that already exist, rather than waiting for custom connectors to be built for each one.

The Renascence take

The interesting part of this story isn't the protocol — it's what it quietly assumes about enterprise readiness. Discovery-based agent architecture only works if the underlying capabilities it "discovers" are well-defined, consistently governed and safe to expose. Most organisations' back-end systems were never designed with that discipline.

Agentic CX will only be as good as the workflow hygiene sitting underneath it — MCP doesn't fix messy data models or inconsistent business logic, it just makes them discoverable at speed. Before enterprises let agents "act on existing capabilities," they should audit which of those capabilities they'd actually want an autonomous system executing unsupervised. The winners here won't be the first to plug in MCP; they'll be the ones who spent the prior year cleaning up what MCP is about to expose.