Salesforce Exposes 200 APIs via MCP for Agent-First Platform Architecture

Salesforce turns platform into discoverable agent fabric

On August 19–20, 2026, Salesforce moved its entire platform to general availability or open beta under a new architecture it calls Headless 360. The core change: nearly 200 Connect API operations and 100+ reusable Agent Skills are now exposed via the Model Context Protocol (MCP), making Salesforce capabilities discoverable and invokable by any MCP-compliant AI agent—whether running in Agentforce, Claude, ChatGPT, Cursor, or other platforms. This is not a product launch. It is a platform bet that every Salesforce capability becomes a governed service callable by AI agents, not just human users.

For enterprise buyers evaluating CRM, commerce, and customer experience platforms, the announcement provides concrete, measurable architecture features to put into RFPs. The Data 360 MCP Server reached GA on August 19, giving agents programmatic access to query endpoints and fully executable actions across the platform. The Headless 360 MCP Server, in open beta since July, acts as the metadata-aware front door for dynamic discovery—agents no longer need developers to manually map every object, API, or workflow before invocation. Salesforce Multi-Framework reached GA, supporting React apps with native authentication, security, and data access. Headless Commerce and the Slackbot MCP Server also hit GA, while the Headless Experience Layer entered open beta, enabling web, mobile, embedded, and conversational UIs on a single logic layer.

Why this changes platform selection and budget planning

The immediate impact is on RFP criteria. Buyers can now require vendors to quantify agent-accessible capabilities. Salesforce's 100+ Agent Skills and exposure of nearly 200 APIs via MCP set a benchmark. Competing platforms that treat AI integrations as per-application add-ons rather than core architecture are now at a measurable disadvantage. Enterprises evaluating SAP, Microsoft Dynamics, or Adobe Experience Cloud should ask: How many capabilities can your platform expose via open standards like MCP? How many agents can discover and invoke them without manual integration work?

Budget and integration planning shift because of the phased rollout. GA components—Data 360 MCP Server, Multi-Framework, Headless Commerce, Slack MCP—can justify near-term production budgets. Open beta components like the Headless 360 MCP Server and Headless Experience Layer are suitable for proof-of-concept or experimental budgets. Because agents dynamically discover capabilities without manual API mapping, integration timelines and costs for AI projects should drop. Buyers can demand vendors quantify this reduction in pilot projects. If a vendor cannot show reduced integration time with agent-driven architecture, the platform is not built for the agentic era.

Governance becomes centralized and observable. By packaging capabilities as MCP tools and Agent Skills, Salesforce centralizes access control and policy enforcement for AI agents. CISOs worried about shadow AI—agents calling SaaS APIs without clear governance—now have a concrete architectural control point. Every agent invocation flows through the MCP Server, where access policies, logging, and compliance checks apply. This matters for industries with strict data governance requirements: financial services, healthcare, public sector.

SAP consolidates BTP into unified AI platform architecture

Around August 21, 2026, SAP renamed SAP Business Technology Platform (SAP BTP) to SAP Business AI Platform (SAP BAIP), consolidating three previously separate pillars—SAP BTP, Business Data Cloud, and Business AI—into one architecture. The move signals a similar strategic shift: converging application, data, and AI layers under a governed platform with a central agent hub and MCP-style server architecture. SAP has not disclosed the same level of quantified capabilities (number of skills, APIs exposed) as Salesforce, but the architectural convergence is material. Enterprises running SAP ERP, S/4HANA, or SuccessFactors now have a unified platform for building and deploying AI agents across business processes, data sources, and extension apps.

The competitive context is clear. Both Salesforce and SAP are betting that SaaS platforms win in an agent-first world by exposing every capability as a governed, discoverable service. Microsoft and Google Cloud have agent frameworks and connector ecosystems, but neither announced comparable quantified capabilities in August 2026. Buyers evaluating Microsoft Power Platform, Azure OpenAI Service, or Google Vertex AI should ask how many enterprise APIs are natively exposed via MCP or similar open standards. The absence of a quantified answer is a red flag.

What to watch: RFP language and pilot metrics

Enterprise buyers should update RFP templates immediately. Add requirements for:

- Number of agent-accessible capabilities (skills, APIs, workflows) exposed via MCP or equivalent open standard - Support for multi-platform agent runtimes (not just the vendor's proprietary agent) - Phased GA vs open beta status of agent architecture components - Centralized governance and observability for agent invocations

For pilot projects, demand measurable integration time reductions. If a platform claims agent-driven architecture but integration still requires manual API mapping, the architecture is marketing, not reality. Track time-to-first-agent-invocation as a benchmark. Salesforce's dynamic discovery via MCP should reduce this from weeks to hours. If a competing platform cannot match that, the agent architecture is not production-ready.

The broader implication: SaaS platforms are splitting into two categories. One category treats AI as an add-on—agents are external consumers of fixed APIs. The other treats AI as the platform's native interface—agents dynamically discover and invoke governed capabilities. The second category will win agent-centric buying decisions. The Salesforce and SAP announcements quantify what that architecture looks like in production.