💡 **MCP is an open standard.** This module shows MCP/Hooks/Skills via Claude Code. MCP servers are also supported by Codex, Hermes, goose, OpenCode, nanobot, and CheetahClaws. Hooks and Skills are shown in Claude Code's format — other stacks have their own equivalents (see module 03).
💭 in your head · nothing to type
Picture a moment when Claude Code stalled mid-task — the right service was right there, but the agent couldn't reach it.
Don't picture this as a bug or a limitation. Picture a craftsman at the workbench: tool belt buckled, but the right socket isn't in the pouch. Hand reaches into an empty slot, work stops. Run it in your head: what exactly was in that empty pocket?
Which tool would you slide in there first?
💭 in your head · nothing to type
Claude Code out of the box is a navigator with precise star charts but no sextant. It knows where to sail, just can't take a fix right here, right now.
Don't think of this as "installing plugins." Picture the navigator finally getting the instrument: one measurement — and position is known. Three mechanisms hand the agent those instruments: MCP (connecting services), Hooks (event automation), Skills (reusable scenarios).
Hold this in your head: which first "measurement" would unlock your work?
| Mechanism | What it does | Analogy |
|---|---|---|
| MCP Server | Gives the agent new tools (GitHub, Notion, browser) | Installing an app |
| Hook | Runs code on events (before/after an agent action) | An automatic trigger |
| Skill | A reusable scenario called via /skill-name |
A macro / script |
This scale is one of the popular author-made frameworks (Avi Chawla, "5 Levels of Agentic AI Systems"), not an accepted standard.
| Level | Type | What it can do |
|---|---|---|
| 1 | Basic Responder | Answers questions |
| 2 | Router | Routes a request to the right agent |
| 3 | Tool-Calling | Uses external tools — Claude Code with MCP |
| 4 | Multi-Agent | Several agents working together |
| 5 | Autonomous | Acts without human involvement |
Claude Code with MCP is level 3. With Skills and parallel agents — level 4.
A level isn't "how much AI replaced you" — it's how much you deliberately delegate. Even at level 5 you hold the frame, the goal, and the check: the tool executes — the role stays yours. You choose which level to climb to for each task.