humanizer MCP — ユースケース, インストール & ライブデモ

なぜ使うのか

主な機能

Pattern library of AI writing tells
Rewrites in your existing style voice when provided
Works on markdown, plain text, and HTML
Can be conservative (minimal edits) or aggressive (full rewrite)
MIT licensed; easy to fork and extend patterns

ライブデモ

実際の動作

準備完了

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": {
    "humanizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/blader/humanizer",
        "~/.claude/skills/humanizer"
      ],
      "_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": {
    "humanizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/blader/humanizer",
        "~/.claude/skills/humanizer"
      ],
      "_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": {
    "humanizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/blader/humanizer",
        "~/.claude/skills/humanizer"
      ],
      "_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": {
    "humanizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/blader/humanizer",
        "~/.claude/skills/humanizer"
      ],
      "_inferred": true
    }
  }
}

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

~/.continue/config.json

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

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

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

claude mcp add humanizer-skill -- git clone https://github.com/blader/humanizer ~/.claude/skills/humanizer

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： humanizer

Strip AI tells from a draft blog post before publishing

👤 Writers who use AI assistance but don't want to sound like everyone else ⏱ ~20 min beginner

使うタイミング： You drafted a post with Claude, and it reads like every LinkedIn thought-leader post from 2025.

前提条件

Skill installed — git clone https://github.com/blader/humanizer ~/.claude/skills/humanizer

フロー

Diagnose

Use humanizer. Analyze /drafts/post.md. List the AI tells — em-dashes, rule-of-three, hedged opener, generic adjectives — with counts.✓ コピーしました

→ Concrete tell inventory, not vibes
Rewrite conservatively

Rewrite paragraph by paragraph. Change only the AI tells. Keep my voice. Show diff.✓ コピーしました

→ Minimal diff, voice preserved
Spot-check

Read the revised version aloud (just the first 200 words). Does it still sound like me?✓ コピーしました

→ Sampled read; flag anything that drifted

結果： A post that doesn't trigger the AI-detection reflex in readers.

注意点

Over-aggressive rewrite flattens your actual voice — Run conservatively first; only escalate sentence-by-sentence on problem areas

組み合わせ： filesystem

Make technical docs sound written, not generated

👤 Teams auto-generating README/API docs ⏱ ~45 min intermediate

使うタイミング： Your auto-generated docs all have "This comprehensive guide will navigate you through..." — and users hate it.

フロー

Scan a batch

Use humanizer on every .md under /docs. For each, report the tell density (tells per 100 words).✓ コピーしました

→ Doc-by-doc density report
Fix worst offenders

Rewrite the 5 with highest density. Stay technical — no vibes, just strip the filler.✓ コピーしました

→ Diffs show removed fluff without losing substance

結果： Docs that read like a human wrote them, which in technical writing means "read like they have something to say".

注意点

Stripping hedging where it was actually a safety note — Review diffs — if a "note that..." was a real warning, keep it

組み合わせ： filesystem

組み合わせ

他のMCPと組み合わせて10倍の力を

humanizer-skill + filesystem

Bulk-humanize a folder of drafts

Run humanizer on every .md under /blog/drafts.✓ コピーしました

humanizer-skill + github

Humanize a PR's auto-generated description

Humanize the description of PR #2341 and push the update.✓ 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
diagnose	text: str	Before rewriting, to see what's wrong
rewrite	text: str, intensity: conservative\|aggressive	Apply the fix
compare_voice	sample_a, sample_b	Check voice preservation after rewrite

コストと制限

運用コスト

APIクォータ: N/A
呼び出しあたりのトークン: Proportional to text length
金額: Free
ヒント: Diagnose first — cheap. Rewrite only what needs it.

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 — skill is prompt-level

トラブルシューティング

よくあるエラーと対処法

Rewrite sounds just as AI as before

You're in conservative mode. Switch to aggressive, and provide 3–5 sample paragraphs of your actual voice as reference.

Skill not activating

Mention it explicitly: "use humanizer". Auto-activation depends on skill metadata.

Voice drifts after rewrite

Pin your voice with a sample: "rewrite to match this paragraph's voice"

代替案

humanizer 他との比較

代替案	代わりに使う場面	トレードオフ
Humanizer-zh	You're writing in Chinese	Chinese-tuned patterns
Manual editing	Short piece; faster to just edit	No systematic pattern catch

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