Section C: Extension Mechanisms

Extension Mechanisms

Five ways to extend Claude Code's capabilities

1
Slash Commands
User-invoked
Custom prompts triggered by /command
2
MCP Servers
External tools
"USB-C for AI": connect to any system
3
Skills
Model-invoked
Claude decides when to use them
4
Subagents
Orchestration
Specialized agents, parallel execution
5
Plugins
Bundles
Package multiple extensions together
Quick Decision Framework
Is it a single, focused capability? ├─ YES: Does Claude need to decide when to use it? │ ├─ YES → Skill │ └─ NO → Slash Command (user controls timing) └─ NO: Do you need parallel execution or context isolation? ├─ YES → Subagents └─ NO → Are you bundling multiple extensions? ├─ YES → Plugin └─ NO → Use MCP Server for external tools