AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Nov 15, 2026, 16:00 JST

MCP vs Skill in Claude Code: how the two differ

Send on LINE
MCP vs Skill in Claude Code: how the two differ

3 Key Points

  1. What happened

    The explainer separates Claude Code extensions: MCP is the common standard connecting AI to external tools and data via MCP servers, Skill is a folder of procedures and scripts led by SKILL.md, and hook enforces rules the system always runs.

  2. Why it matters

    Treating them as competing misses the point — MCP delivers the connection, Skill delivers the procedure, and hook guarantees enforcement, so users can pick which they lack.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The explainer builds its comparison around a simple analogy: a capable new hire who lacks both account access and a sense of how the team works. Handing over accounts maps to MCP, and handing over a handover document maps to Skill. That framing carries through the rest of the piece, where MCP is described as a common standard sitting between Claude and external systems, with an MCP server acting as an adapter for each system. Skill is described as a reusable folder whose only required file is SKILL.md, with a name and description at the top and procedures written in the body.

The piece also explains why many Skills do not weigh Claude down: at startup only each Skill's description is read, the body loads when a request matches, and references or scripts open only as needed — compared to scanning a library's spines but opening only the books you need. MCP and Skill are then presented as complements rather than rivals, with the choice resting on what is missing: outside access points to MCP, a way of working points to Skill, and rules that must always run point to hook.

FAQ
How is Skill different from CLAUDE.md?
CLAUDE.md is a short note read on every conversation, while Skill loads its body only when a request matches its description. Long or occasional procedures belong in Skill; short rules meant every time go in CLAUDE.md.
Does connecting many MCP servers slow Claude Code down?
Tool search looks up tools only when needed, which keeps startup load low. Even so, the basic practice is to disconnect servers you do not use.
Can I build my own MCP server?
Yes. An official SDK is available, and wrapping an internal API as an MCP server makes it usable from Claude Code. It is worth first checking whether an existing server is enough.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleLlama 3.2 1B on QCS6490: 1,024 tok/s prefill, 16.4 tok/s decode