Salesforce's Headless 360 initiative lets developers expose custom Apex business logic as Hosted MCP (Model Context Protocol) tools that AI agents like Claude or Cursor can discover and invoke directly — no UI required. The post walks through writing an Apex @InvocableMethod for pipeline intelligence, registering it as an MCP Server via metadata, configuring OAuth via an External Client App, and connecting Claude to query Salesforce data conversationally. Key design guidance covers writing agent-readable tool descriptions, using computed fields instead of raw data, fuzzy account matching, and enforcing record-level security with USER_MODE SOQL.

10m read timeFrom developer.salesforce.com
Post cover image
Table of contents
What makes Salesforce Hosted MCP differentExample: Pipeline intelligenceRegistering the Hosted MCP ServerSetting up the External Client AppConnecting ClaudeThe Headless 360 shiftA note on tool governanceGet startedResourcesAbout the author

Sort: