Salesforce Headless 360 and MCP
For most of Salesforce’s history, building on the platform meant working inside it: browser open, Setup menu navigated, page layouts configured. Salesforce Headless 360, announced at TDX on April 15, draws a line under that era. Every piece of the platform (data, workflows, business logic, compliance controls) is now reachable as an API, an MCP tool, or a CLI command. No browser required. Here is what that actually means if you build integrations or work with AI coding tools. Salesforce Headless 360 — How the Platform Is Now Exposed Salesforce Platform Data · Flows · Logic · Permissions Headless 360 REST / Bulk / Streaming Standard Salesforce APIs 60+ MCP Tools AI-callable structured tools CLI / DevOps Center Programmatic CI/CD access Human Developers CLI · VS Code · CI/CD pipelines AI Coding Agents Claude Code · Cursor · Codex Integration Systems MCP servers · External APIs No browser required. Every capability reachable as API, MCP tool, or CLI command. What Salesforce Headless 360 Ships With The headline number is 60 new MCP tools exposing Salesforce capabilities as structured interfaces that AI agents can call directly: objects, flows, deployments, permissions, reports. Alongside those, 30 preconfigured coding skills cover the most common Salesforce development tasks, from writing Apex triggers to configuring permission sets. DevOps Center MCP and Natural Language DevOps The DevOps Center MCP is the most immediately practical piece for integration teams: it brings programmatic access into any CI/CD pipeline. Natural Language DevOps sits on top of that: describe what you want to deploy in plain language and agents execute it. Salesforce cites a 40 percent cycle time reduction for development workflows as an early benchmark from teams using the integrated toolchain. The framing from Salesforce is that the build loop requiring context-switching across four different tools now happens inside one connected experience. That is the right framing, but the practical reality depends heavily on where your org’s metadata and documentation quality sits right now. What MCP means for Salesforce integrations The Model Context Protocol is becoming the default interoperability layer for AI agents across the industry. By exposing Salesforce through MCP, Salesforce is positioning itself as a participant in multi-agent orchestration, not just a destination for data, but a system that AI agents from Claude Code, Cursor, Codex, and Windsurf can work within directly. The practical shift for integration teams is significant. Instead of writing API wrappers manually (defining the request shape, handling authentication, mapping the response), you describe the integration behaviour and an agent executes and tests it using the MCP tools that already understand Salesforce’s object model. A Practical Integration Example For example: an integration that syncs Salesforce Opportunity data to an external billing system previously required a developer to write and maintain the API connection, handle field mapping, and manage error states. With MCP-exposed Salesforce tools and an AI coding agent, the same integration can be described in plain language, generated, and validated against the actual org schema, including custom objects and fields, without manual API documentation reading. The constraint is the quality of the description. Vague integration requirements produce vague integrations. MCP tools remove the API plumbing problem; they do not remove the requirement to know what you are building. Agent Fabric and multi-agent governance Headless 360 opens the Salesforce platform to AI agents. Agent Fabric answers the follow-on question: when you have multiple agents from multiple vendors operating across Salesforce and other systems, who governs them? Agent Fabric is the control plane launched at TDX 2026. It provides deterministic orchestration and centralised governance across agents, tools, and LLMs, scanning for agents across Salesforce Agentforce, Amazon Bedrock, Microsoft Foundry, and MCP servers, then providing a unified view of what agents exist, what they have access to, and what they are doing. Why Deterministic Governance Matters The governance problem it solves is specific: AI agents are probabilistic. They reason toward outcomes, which means their behaviour cannot be fully predicted from the instruction set alone. Agent Fabric adds the deterministic layer: Agent Broker defines fixed handoff rules between agents, Agent Script enforces compliance checkpoints within workflows, and Trusted Agent Identity ensures agents execute with specific user permissions rather than a shared service account. For architects designing systems where agents span Salesforce and external platforms, Agent Fabric is the piece that makes enterprise-grade deployment possible rather than merely experimental. Workflow Traditional Salesforce dev Headless 360 + AI coding agents Write an Apex trigger Open Developer Console or VS Code → write Apex → deploy to sandbox → test manually → deploy to production Describe the trigger behaviour in natural language → agent writes Apex using your org’s actual object model → review → deploy via DevOps Center MCP Build a Salesforce integration Read API docs → write request/response handlers → map fields manually → handle auth and error states → test in sandbox Describe integration behaviour → agent generates using MCP tools that understand your schema → automated testing → review and approve Understand an existing automation Open Flow Builder or Apex class → read through the logic → trace dependencies manually → ask a colleague Ask Vibes (Ask mode): “Explain this trigger and whether it conflicts with the Process Builder on this object” — org-aware answer with dependency context Debug a production issue Reproduce locally → inspect error log → trace through code → write fix → test → deploy Debug mode: agent analyses error log, identifies root cause in org context, proposes targeted fix → developer reviews and applies Deploy a change across environments Package the change → run change set or Salesforce CLI deploy → validate in staging → release Natural Language DevOps: describe the deployment in plain language → agent executes via DevOps Center MCP with CI/CD pipeline integration Context-switching across tools VS Code → Org → Developer Console → Sandbox → Jira → Slack → back to Org. Multiple context switches per task. One connected experience inside the development environment. Salesforce cites 40% reduction in development cycle time. What you need before Headless
