Salesforce introduces its first official Claude Code plugin, packaging around 40 development skills from its open-source sf-skills library, three MCP servers (including an Apex/SOQL language server), specialized agents, hooks, and slash commands into a single install. The post explains the building blocks of headless development (agent skills, MCP servers, agent plugins), how Salesforce restructured its skills library with a naming taxonomy and per-skill versioning for governance at scale, and walks through setup steps requiring Claude Code, Node.js, Salesforce CLI, and Python. It closes with a roadmap covering a Claude Cowork plugin, open-source plugin architecture, and governance tooling (CAP-G).
Table of contents
The components of headless developmentHow Salesforce is scaling headless developmentWhat’s nextConclusionResourcesAbout the authorQuestions this post answers
What does the Salesforce development plugin for Claude Code include?
It bundles about 40 Salesforce development skills, three MCP servers, specialized agents, hooks, and slash commands into a single installable unit. The MCP servers include salesforce-api-context for API/metadata guidance, salesforce-metadata-experts for metadata type knowledge, and salesforce-lsp which hosts local Apex and SOQL language servers for real-time diagnostics and completions. daily.dev surfaces plugin ecosystem updates for developers wiring AI agents into their Salesforce workflow.
What are the prerequisites for installing the Salesforce development plugin for Claude Code?
You need Claude Code, Node.js LTS (v22 or v24), Salesforce CLI, and Python 3.8 or above. Python is only required for internal hooks related to org detection and deployment safety, and that requirement may be removed in future versions. After installing, you authenticate to an org via Salesforce CLI and add the plugin through the Claude Plugin Marketplace. Developers setting up new agent tooling can track prerequisite changes like this on daily.dev.
What is the difference between agent skills, MCP servers, and agent plugins in headless AI development?
Agent skills are reusable folders containing a SKILL.md file with instructions that agents lazy-load into context as needed; MCP servers use the open Model Context Protocol to let agents connect to external system APIs in a standardized, plug-and-play way; agent plugins package skills, MCP servers, and other utilities into one installable, updatable unit, though plugins remain vendor-specific unlike the open skill and MCP standards. Anyone piecing together an agent stack can compare these building blocks on daily.dev before choosing a setup.