Salesforce has moved its most ambitious agentic bet into public testing, opening the Headless 360 MCP Server in beta in early July 2026. The release gives autonomous AI agents a stable, standardized doorway into the company's entire platform — data, workflows, Apex code and configuration — without a human ever opening a browser. It marks a concrete step in Salesforce's multi-year effort to rebuild its software around agents rather than clicks.
From Browser Clicks to Agent Calls
Headless 360 was first unveiled at the TrailblazerDX conference earlier this year as a sweeping reorientation of the platform. The core idea is to expose every Salesforce capability — CRM, Agentforce, Data Cloud and Slack — as APIs, Model Context Protocol (MCP) tools and command-line operations that software agents can call directly. "We made a decision two and a half years ago: Rebuild Salesforce for agents," the company said in announcing the initiative.
The July beta builds on hosted MCP servers that reached general availability in April. What makes the new server notable is not scale but restraint. Rather than flooding an agent with thousands of individual tools, Salesforce designed the server to present just four tools backed by a continuously growing library of skills. The agent's surface stays small and predictable, while the set of actions it can perform expands independently behind that facade.
How the Four-Tool Design Works
The tools mirror how an agent actually processes a request. A Discover tool runs a semantic search across a vector index of every generated API and skill, returning a ranked shortlist of candidates so the agent spends less time hunting for the right capability. A Describe tool then pulls the technical contract for a chosen skill — its parameters, dependencies and the ordered steps required to run it. A Dispatch tool executes the underlying API calls.
This architecture directly addresses a growing pain point in agentic engineering: tool sprawl. When an agent is handed hundreds or thousands of raw endpoints, it wastes tokens and makes more mistakes simply trying to locate the correct one. By collapsing discovery, description and execution into a tight loop, Salesforce aims to keep agents fast and accurate even as the underlying platform grows.
For developers, the practical payoff is direct. Coding agents such as Claude Code, Cursor, Codex and Windsurf can query Salesforce data, run SOQL, trigger workflows and invoke Apex without a graphical interface. A developer can instruct an agent to run every Apex test for a given class and surface the failures, and it executes that end to end.
New July Capabilities
The beta expands into higher-stakes administrative territory. Newly supported tasks include managing users — creating and deactivating accounts, freezing and unfreezing them, resetting passwords and assigning permission sets — as well as reading, writing and deploying Apex triggers, building event-driven integrations and creating named credentials.
To keep pace, Salesforce built an internal system it calls the Headless 360 Factory, which scans Setup, identifies features that lack API access and automatically generates the APIs and skills that agents can then use through the MCP server. That factory model is how the company intends to close coverage gaps faster than it could by hand.
Why It Matters
Headless 360 is a signal of where enterprise software is heading. Salesforce is repositioning itself from a vendor that runs "AI agents inside Salesforce" to a programmable platform for agents that operate across external tools and environments — including Slack, Microsoft Teams, and third-party assistants that support MCP. Company leaders frame it as preparation for a future in which most code is written by agents, a shift they say is already visible in their own internal systems.
But the beta also sharpens a debate about governance. Analysts note that Salesforce has built a powerful access layer without a matching governance layer. The tooling includes no native quality checks on agent-generated code, no automatic assessment of whether new Apex introduces technical debt or violates naming conventions. Handing coding agents live keys to a production org raises real questions about who reviews what those agents build.
Cost and capacity add further caution. Enterprise Edition orgs start with a ceiling of roughly 100,000 REST API calls in a rolling 24-hour window, pooled across API types — a limit that arrives quickly when agents poll the platform every few seconds. Analysts also warn that new capabilities may land behind premium licenses, and advise buyers to press on pricing before building architectural dependencies.
The Bigger Picture
The Headless 360 MCP Server lands amid a broader July surge in agentic tooling, as the industry pivots from flashy demos toward agents that reliably replace real workflows. Salesforce's bet is that the winners will be platforms that make themselves natively legible to agents — discoverable, callable and safe at machine speed. The beta is a test of whether a small, disciplined tool surface can deliver that promise without sacrificing the guardrails enterprises still need.
