01
Overview
The Salesforce Headless Toolkit is the developer-facing architecture underneath AIforce. Its purpose is to separate Salesforce capabilities from a fixed browser UI so agents, models and third-party systems can invoke CRM data and actions through machine-friendly interfaces.
Salesforce currently documents four important access patterns: Model Context Protocol (MCP), APIs, CLI commands and Skills. They solve related but different problems. MCP gives compatible AI clients a standardized way to discover and call Salesforce capabilities. APIs remain the programmatic interface for applications that need explicit integration logic. CLI commands support developer and administrative workflows. Skills package reusable capabilities that an AI system can load and execute.
02
MCP access
The key architectural point is governance. AIforce is not presented as a separate copy of Salesforce data. Salesforce says the Headless Toolkit exposes the existing platform architecture, including permissions, business logic, semantics, security and governance. That means an AI interface can reason over Salesforce context while actions still flow through the organization's Salesforce rules.
A concrete example is Salesforce in Claude. Salesforce says it packages Salesforce as a prebuilt MCP server inside Claude, avoiding manual MCP setup, custom authentication and skill mapping for the initial experience. It launches with 37 prebuilt sales skills. ProjectMonet.space already has a separate Claudeforce canonical, so this guide focuses on the underlying developer architecture rather than repeating the sales-product overview.
03
APIs and CLI
For custom builders, the Headless Toolkit broadens the possible interface. Salesforce says developers can create agentic experiences in AI tools and IDEs rather than forcing users back into Salesforce pages. The launch announcement also describes AgentExchange as an ecosystem where partners can build interfaces, agents, apps, integrations, workflows and actions on this architecture.
04
Skills and development workflow
MCP should not be interpreted as bypassing authorization. Salesforce's positioning is the opposite: the headless interfaces are meant to preserve existing permissions and governance. Developers still need to evaluate the exact authentication flow, scopes, available tools and organization policies for their chosen surface.
The same caution applies to Zero Data Retention. Salesforce says AIforce can use model providers under ZDR terms so business data used to answer a request is not retained by the model provider. That does not automatically describe every log, integration, external tool or customer configuration. Treat it as a vendor claim that needs deployment-specific verification.
05
Choosing the right interface
A practical evaluation checklist for developers: identify the AI surface you want to use; determine whether a prebuilt Salesforce integration exists; map the Salesforce data and actions required; inspect existing permissions and business rules; choose MCP, API, CLI or Skills based on the workflow; verify model-provider and retention requirements; test write actions with human controls where appropriate; and validate auditability before production rollout.
AIforce is new, so exact endpoints, packaging and supported third-party surfaces can evolve. Use current Salesforce developer documentation before implementing production code rather than relying on launch-day examples as permanent API contracts.
06
Limits and verification
FAQ: Is the Headless Toolkit an API? It is broader than one API; Salesforce describes it as an architecture/toolkit spanning MCP, APIs, CLI commands, Skills and developer tools. Can AIforce work outside Salesforce's UI? Yes—that is the core headless premise. Does Salesforce in Claude require manually building an MCP server? Salesforce says its beta includes a prebuilt MCP server and centralized setup. Does MCP remove Salesforce permissions? No; Salesforce says existing permissions and business rules continue to govern access and actions.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.