Salesforce Headless 360 Explained: What 60 MCP Tools Actually Mean for Your Org

Salesforce Headless 360 is the most significant platform architecture announcement since Lightning. That's not hyperbole — it's a structural change to how software accesses Salesforce. Understanding what it actually is (versus what the marketing says) matters for every architect making decisions about their org this year.

What Headless 360 actually is

The announcement: every major Salesforce capability is now accessible as a REST API, a Model Context Protocol (MCP) tool, or a CLI command. No browser session. No Lightning UI. No manual click-through. That means AI agents — Agentforce, Claude Code, Cursor, Codex, Windsurf, your own custom agents — can now query org data, trigger workflows, deploy metadata, and run business logic by calling these surfaces directly.

The surface is substantial. 60+ new MCP tools. 30+ preconfigured coding skills. 4,000+ existing APIs. 220+ CLI commands. Salesforce Hosted MCP Servers for the core platform, Tableau Next, Data 360 SQL, and MuleSoft went GA in April 2026 — no local setup required, natively integrated, governed by Salesforce's own trust layer.

The security model you must understand before anything else

Before jumping to 'what do I build', the security model deserves careful attention. Four enforcement mechanisms apply universally to every MCP call:

• Identity verification — the agent acts as a specific, authenticated user
• Access control — sharing rules and field-level security remain in force
• Invocation scope — you explicitly define which tools are exposed
• Policy enforcement — validation rules, triggers, approval chains, and governor limits all still fire

Headless 360 does not bypass your security model. It applies it headlessly. That distinction matters when explaining this to compliance teams.

The Data 360 facade pattern — the design worth studying

The Data 360 MCP Server (Developer Preview, May 2026) is worth examining for its design approach. Rather than exposing approximately 200 API operations individually — which would overwhelm any LLM context window — Salesforce consolidated them behind three facade tools. That facade architecture is how enterprise-scale AI data access gets solved. It's the pattern every architect building custom integrations should be learning from and applying to their own metadata design.

Agentforce Vibes 2.0 — test it this week

For developers, Agentforce Vibes 2.0 is the immediate action item. It's free in Developer Edition orgs, uses Claude Sonnet as default (with GPT-5 and Salesforce models available), and claims 40% faster build cycles. Treat that figure sceptically until you've run your own benchmarks — but the tool is free, and not testing it is a choice to remain uninformed. Stand up a Developer Edition org today if you haven't.

Multi-framework support changes the ISV calculation

React can now run natively on the Salesforce platform. For ISVs and internal teams that have avoided deep Salesforce customisation because of LWC friction or Aura legacy complexity, this reshapes the build-vs-buy calculus. The Agentforce Experience Layer (AXL) means one agent or app definition renders correctly across Slack, Salesforce Mobile, ChatGPT, Teams, and any MCP-compatible client — one definition, all surfaces, no surface-specific rework.

What hasn't changed — and why this matters

Headless 360 runs alongside Lightning, not instead of it. Your existing flows, configurations, and custom components remain intact. The headless layer is additive. Organisations that interpret this announcement as 'we need to rebuild everything' are misreading it. The correct interpretation is 'we now have an additional architectural surface to govern and exploit deliberately'.

The 300% growth in AI agents on Slack since January 2026 and 13,000+ listings in the new AgentExchange Marketplace tell you where the ecosystem is moving. The architectural work — governance standards for agent definitions, MCP tool exposure policies, Data 360 integration patterns — needs to happen now, before the ecosystem fragments into ungoverned proliferation. The Agentforce Script language going open source accelerates this timeline. Community-built agent definitions will proliferate fast. Establish your internal governance standards before you need them.