CCA-F: Claude Certified Architect — Foundations

MCP servers can expose prompts as one of the three MCP primitive types (tools, resources, and prompts). In Claude Code, MCP prompts are surfaced as slash commands using the naming convention /mcp__servername__promptname — for example, /mcp__devops__deploy_checklist — and any parameters the prompt defines are supplied as arguments typed after the command name, much like arguments passed to custom slash commands. This is distinct from MCP resources, which are instead surfaced as @-mentionable items that get fetched and attached to a message, and distinct from MCP tools, which populate the tool registry and are invoked automatically by the model during a task. Because prompts are reusable, explicitly user-triggered templates (not autonomous tool calls or injected system context), Claude Code exposes them through the slash-command interface rather than auto-invoking or auto-injecting them.