skills MCP — ユースケース, インストール & ライブデモ

Name: skills MCP Server
Author: matlab

なぜ使うのか

主な機能

Official: maintained by MathWorks
Covers core MATLAB language plus common toolboxes
Encodes MATLAB Projects and unit-test conventions
Performance guidance: vectorization, preallocation, profiling
Works with Claude Code and other skill-compatible agents

ライブデモ

実際の動作

skills-skill-4.replay ▶ 準備完了

0/0

インストール

クライアントを選択

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "skills-skill-4": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/matlab/skills",
        "~/.claude/skills/skills"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config を開く。保存後、アプリを再起動。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "skills-skill-4": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/matlab/skills",
        "~/.claude/skills/skills"
      ],
      "_inferred": true
    }
  }
}

Cursor は Claude Desktop と同じ mcpServers スキーマを使用。プロジェクト設定はグローバルより優先。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "skills-skill-4": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/matlab/skills",
        "~/.claude/skills/skills"
      ],
      "_inferred": true
    }
  }
}

Cline サイドバーの MCP Servers アイコンをクリックし、"Edit Configuration" を選択。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "skills-skill-4": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/matlab/skills",
        "~/.claude/skills/skills"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop と同じ形式。Windsurf を再起動して反映。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "skills-skill-4",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/matlab/skills",
        "~/.claude/skills/skills"
      ]
    }
  ]
}

Continue はマップではなくサーバーオブジェクトの配列を使用。

~/.config/zed/settings.json

{
  "context_servers": {
    "skills-skill-4": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/matlab/skills",
          "~/.claude/skills/skills"
        ]
      }
    }
  }
}

context_servers に追加。保存時に Zed がホットリロード。

claude mcp add skills-skill-4 -- git clone https://github.com/matlab/skills ~/.claude/skills/skills

ワンライナー。claude mcp list で確認、claude mcp remove で削除。

ユースケース

実用的な使い方： skills

How to get Claude to write idiomatic MATLAB, not MATLAB-flavored Python

👤 MATLAB developers who noticed Claude defaulting to loops everywhere ⏱ ~25 min intermediate

使うタイミング： Any MATLAB authoring task — from scripts to toolboxes.

前提条件

MATLAB R2023b+ recommended — mathworks.com/products/matlab.html
Skill cloned — git clone https://github.com/matlab/skills ~/.claude/skills/skills

フロー

Describe the task

Write a function that computes the moving-window correlation of two time series — MATLAB, idiomatic.✓ コピーしました

→ Vectorized implementation, correct function signature, docstring
Ask for tests

Add MATLAB unit tests using matlab.unittest.✓ コピーしました

→ Tests in a separate file following MATLAB Test conventions
Profile

If this is called in a tight loop, how would you profile and speed it up?✓ コピーしました

→ tic/toc or profile() suggestions, preallocation hints

結果： Idiomatic, fast, tested MATLAB code.

注意点

Claude reaches for Python-style loops — Explicitly ask for vectorized form; reference MATLAB docs style

組み合わせ： filesystem

Port a Python/NumPy script to MATLAB correctly

👤 Engineers moving research code into a MATLAB pipeline ⏱ ~30 min intermediate

使うタイミング： You have a NumPy prototype and need a clean MATLAB equivalent.

フロー

Share the Python

Port this NumPy script to MATLAB. Preserve numerical behavior — flag where semantics differ.✓ コピーしました

→ Translated code + explicit notes on 1-indexing, broadcasting, default dtypes
Verify with a test

Write a side-by-side test that shows both produce the same outputs on known inputs.✓ コピーしました

→ A small cross-check script

結果： MATLAB port with documented semantic differences.

注意点

Off-by-one from zero- vs one-indexing — Claude should call these out; if not, re-ask

組み合わせ

他のMCPと組み合わせて10倍の力を

skills-skill-4 + filesystem

Work across a MATLAB Projects folder structure

Read +mytools/, add a new function, and update the tests package.✓ コピーしました

skills-skill-4 + github

Open PRs for MATLAB codebases with proper test coverage

Open a PR that adds the new function + tests.✓ コピーしました

ツール

このMCPが提供する機能

ツール	入力	呼び出すタイミング
matlab-authoring	task description	Any MATLAB task
matlab-testing	function under test	After authoring
perf-guidance	slow code + context	When something is too slow

コストと制限

運用コスト

APIクォータ: none
呼び出しあたりのトークン: ~3–10k per task
金額: free; MATLAB license sold separately
ヒント: Constrain to a specific toolbox when asking — reduces context

セキュリティ

権限、シークレット、影響範囲

認証情報の保管： none

データ送信先： none

トラブルシューティング

よくあるエラーと対処法

Generated code uses a toolbox you don't have licensed

Tell Claude which toolboxes are available.

確認： ver in MATLAB

Script errors in older MATLAB

Specify your MATLAB release; skill adapts suggestions.

確認： version in MATLAB

代替案

skills 他との比較

代替案	代わりに使う場面	トレードオフ
jupyter	You're really doing Python science and only occasionally MATLAB	Different ecosystem

その他

リソース

📖 GitHub の公式 README を読む

🐙 オープンな issue を見る

🔍 400以上のMCPサーバーとSkillsを見る