Salesforce Inc. spent 27 years persuading enterprises that its user interface was the product. On Wednesday it made the opposite case. The company pushed its Data 360 MCP Server to general availability, exposing nearly 200 customer-data APIs through the Model Context Protocol so that any authorized AI agent — Anthropic's Claude, OpenAI's ChatGPT, Cursor, or Salesforce's own Agentforce — can discover, query, transform and activate governed customer data without a human ever opening a browser tab.
The move, reported Wednesday by SiliconANGLE, is the second act of Headless 360, the architecture Salesforce unveiled at TrailblazerDX in April under the slogan "our API is the UI." That first release turned roughly 60 platform capabilities into MCP tools so coding agents could reach Salesforce workflows. Data 360 goes a layer deeper, into the data itself.
Until now, headless access to Data 360 mostly meant reading. The new server lets agents write: build semantic models, generate calculated insights, suggest field mappings across systems, inspect identity graphs down to individual records, create audience segments, activate campaigns and stand up entirely new data streams — all from natural language. Salesforce's own example: a business user types "lifetime value of all customers for electronics purchases, excluding software," and the agent assembles the relationships, transforms and formulas before running the query. Work that took weeks of coordination between data engineers and business teams, the company argues, collapses to minutes.
Wednesday's announcement bundled more than the data server. The Headless 360 MCP Server entered open beta; Slackbot's MCP client went generally available with connections to more than 20 partner apps including Atlassian, Box, Docusign, Notion and Zoom; and a repository of more than 100 reusable Skills shipped. Prebuilt Data 360 Skills for modeling, mapping, transforms and activation are due later this month.
The numbers behind the pitch are the reason to take it seriously. Data 360 ingested 32 trillion records in Salesforce's third quarter of fiscal 2026, up 119% year over year, including 15 trillion through zero-copy connections, up 341%. In the first quarter of fiscal 2027, reported in May, Agentforce crossed $1 billion in annual recurring revenue, up 205%, with combined AI and data ARR at $3.4 billion on $11.13 billion in quarterly revenue. Customers running Agentforce in production grew 50% in a single quarter, and Salesforce said it delivered 28.6 trillion tokens, up 152% sequentially, across 3.8 billion agentic work units.
"Data 360 Headless lets our agents reach a unified patient context from any surface — through MCP, without waiting for a purpose-built interface for each one," said Dr. Paramjit "Romi" Chopra, founder and chief executive of MIMIT Health, an early customer. "What makes those answers clinically trustworthy is the physician-curated ontology sitting between the data and the agent. That's where the intelligence lives." It is the whole thesis, stated by a customer: the interface is disposable, the curated context is not. Salesforce co-founder Parker Harris framed it more bluntly this spring, asking why anyone should ever log into Salesforce again.
Investors are unconvinced. CRM shares are down roughly 34% in 2026 against an S&P 500 up nearly 9%, on the fear that agents will quietly retire the human seats Salesforce has billed for since 1999. Bank of America's Tal Liani reinstated an underperform rating with a $160 target in May, calling it an AI-driven structural reset. Bulls counter with $72.4 billion in remaining performance obligations and a forward EV/EBITDA multiple near 9.7x, against a peer group closer to 29x. Both camps agree the seat model is under pressure; they disagree on whether consumption revenue scales fast enough to replace it.
Why It Matters
MCP has become the connective tissue of enterprise agents faster than almost anyone predicted. Anthropic open-sourced the protocol in November 2024; OpenAI, Google, Microsoft, IBM and Amazon adopted it. In December 2025 Anthropic donated it to the Linux Foundation's Agentic AI Foundation, backed by AWS, Google, Microsoft, OpenAI, Bloomberg and Cloudflare; membership hit 146 by late February. The SDK reached 97 million monthly downloads by March, and roughly 28% of Fortune 500 companies now run MCP servers in production.
For SaaS vendors, that is an uncomfortable gift. A universal tool-calling standard turns every application's capabilities into substitutable line items inside someone else's agent loop. The UI — the thing SaaS companies spent two decades polishing, training users on and quietly relying on for renewal stickiness — stops being where the work happens. Salesforce's answer is to embrace the disintermediation and relocate the moat: if agents can reach anything, the durable asset is unified, permissioned, semantically modeled customer data that inherits 25 years of business logic, sharing rules and compliance controls. Model intelligence without that context, as the company's April framing put it, is just inference.
The bet is not risk-free. Making capabilities addressable from Claude or ChatGPT means Salesforce is no longer the destination, and a customer whose agents pull context through MCP may find it easier, not harder, to swap the application layer above that data. Salesforce is wagering that governance is stickier than screens.
What To Watch
Three things. First, Data 360 Skills GA later in August — prepackaged workflows are where the hours-not-months claim meets real enterprise data hygiene. Second, second-quarter fiscal 2027 earnings in early September: whether Data 360 consumption revenue accelerates fast enough to answer the seat-erosion bears, and whether Salesforce discloses MCP-driven usage at all. Third, the direction of traffic. If enterprises route third-party agents through Salesforce's MCP servers, the data-as-moat thesis holds. If they instead point those agents at Snowflake, Databricks or their own lakehouses and treat Salesforce as one endpoint among many, the headless strategy will have opened a door the company cannot close. Security is the sleeper variable: the NSA published MCP hardening guidance in June, and prompt-injection exposure scales with every API exposed.
“Data 360 Headless lets our agents reach a unified patient context from any surface, through MCP, without waiting for a purpose-built interface for each one.”— Dr. Paramjit Chopra, Founder and CEO, MIMIT Health