@mako10k/mcp-confirm
AI-ユーザー間の復唱確認プロトコルを実装するMCPサーバーです。LLMが不安になったときに、ユーザーに確認を取るためのツールを提供します。
@mako10k/mcp-confirm
A Model Context Protocol (MCP) server for AI-user confirmation and clarification. This server provides tools for AI assistants (LLMs) to ask users for confirmation when they need clarification or verification.
Overview
This MCP server implements the Model Context Protocol Elicitation specification to enable confirmation and clarification protocols between AI assistants and users.
AI assistants can use this in situations such as:
- Confirming actions before execution
- Clarifying ambiguous requests
- Verifying understanding is correct
- Asking yes/no questions
- Collecting user satisfaction ratings
Features
- ask_yes_no
- Ask yes/no confirmation questions
- Used when AI needs clarification or verification
- confirm_action
- Confirm actions before execution
- Includes impact and details in confirmation dialog
- clarify_intent
- Clarify ambiguous requests
- Present multiple interpretation options
- verify_understanding
- Verify AI's understanding is correct
- Confirm next steps before proceeding
- collect_rating
- Collect user satisfaction ratings
- Evaluate AI response quality
- elicit_custom
- Custom confirmation dialogs
- Use custom JSON schemas
- search_logs
- Search confirmation history logs
- Filter by type, success status, date range, response time
- Paginated results for large datasets
- analyze_logs
- Statistical analysis of confirmation history
- Success rates, response times, trends
- Grouping by type, time period, etc.
Installation
# Install globally
npm install -g @mako10k/mcp-confirm
# Or use with npx
npx @mako10k/mcp-confirm
Configuration
Environment Variables
You can configure the server using environment variables:
MCP_CONFIRM_LOG_PATH: Path to confirmation history log file (default: .mcp-data/confirmation_history.log)MCP_CONFIRM_TIMEOUT_MS: Default timeout for confirmations in milliseconds (default: 60000)NODE_ENV: Set to development to enable debug logging
Timeout Behavior
The server uses intelligent timeout settings based on confirmation type:
- Critical actions (delete, remove operations): 120 seconds
- Warning actions: 90 seconds
- Simple yes/no questions: 30 seconds
- Rating requests: 20 seconds (reference only)
- Other confirmations: 60 seconds (default)
Confirmation History Logging
All confirmation interactions are logged to a file for audit purposes. The log includes:
- Timestamp of the request
- Confirmation type
- Full request and response data
- Response time in milliseconds
- Success/failure status
- Error messages (if any)
The log directory (.mcp-data/) will be created automatically if it doesn't exist.
VS Code Integration
Add to your .vscode/mcp.json:
{
"servers": {
"mcp-confirm": {
"type": "stdio",
"command": "npx",
"args": ["@mako10k/mcp-confirm"]
}
}
}
Claude Desktop Configuration
Add to your Claude Desktop config.json:
Windows
Location: %APPDATA%\Claude\config.json
macOS
Location: ~/Library/Application Support/Claude/config.json
Linux
Location: ~/.config/claude/config.json
{
"mcp": {
"servers": {
"mcp-confirm": {
"command": "npx",
"args": ["@mako10k/mcp-confirm"],
"env": {
"MCP_CONFIRM_LOG_PATH": "~/.mcp-data/confirmation_history.log",
"MCP_CONFIRM_TIMEOUT_MS": "60000",
"NODE_ENV": "production"
}
}
}
}
}
Usage Examples
Basic Confirmation
// Before AI executes an action
await confirm_action({
action: "Delete files",
impact: "Cannot be restored",
details: "10 files will be deleted"
});
Intent Clarification
// When request is ambiguous
await clarify_intent({
request_summary: "Want to create a project",
ambiguity: "Type of project unclear",
options: ["Node.js project", "Python project", "React app"]
});
Understanding Verification
// Before complex tasks
await verify_understanding({
understanding: "Create web application with user authentication",
key_points: ["React + Node.js", "JWT authentication", "PostgreSQL database"],
next_steps: "Create project structure then implement auth system"
});
Technical Specifications
- Protocol: Model Context Protocol Elicitation
- Language: TypeScript
- Runtime: Node.js
- SDK Version: @modelcontextprotocol/sdk ^1.0.0
How It Works
This server implements true MCP Elicitation protocol:
- Sends
elicitation/create method to client - Defines user input structure with JSON Schema
- User responds with
accept, decline, or cancel - On
accept, receives structured data following schema
This enables reliable communication between AI and users.
Development
# Clone repository
git clone https://github.com/mako10k/mcp-confirm.git
cd mcp-confirm
# Install dependencies
npm install
# Build
npm run build
# Run locally
npm start
# Quality checks
npm run quality
License
MIT License - see LICENSE file for details.
Repository
機能
利用可能なツール
- ask_yes_no
- はい/いいえの確認質問
- AIが明確化や検証が必要な時に使用
- confirm_action
- アクション実行前の確認
- 影響や詳細を含めた確認ダイアログ
- clarify_intent
- あいまいなリクエストの意図確認
- 複数の解釈オプションの提示
- verify_understanding
- collect_rating
- elicit_custom
- カスタム確認ダイアログ
- 独自のJSONスキーマを使用
セットアップ
# 依存関係のインストール
npm install
# ビルド
npm run build
# 実行（グローバルインストール後）
npm install -g @mako10k/mcp-confirm
mcp-confirm
# または npx で直接実行
npx @mako10k/mcp-confirm
VS Code統合
.vscode/mcp.jsonでMCPクライアントとして設定済み：
{
"servers": {
"mcp-confirm": {
"type": "stdio",
"command": "npx",
"args": ["@mako10k/mcp-confirm"]
}
}
}
使用例
基本的な確認
// AIがアクションを実行する前
await confirm_action({
action: "ファイルを削除",
impact: "復元できません",
details: "10個のファイルが削除されます"
});
意図の確認
// あいまいなリクエストの場合
await clarify_intent({
request_summary: "プロジェクトを作成したい",
ambiguity: "どの種類のプロジェクトか不明",
options: ["Node.js プロジェクト", "Python プロジェクト", "React アプリ"]
});
理解の検証
// 複雑なタスクの前
await verify_understanding({
understanding: "Webアプリケーションを作成して、ユーザー認証を実装する",
key_points: ["React + Node.js", "JWT認証", "PostgreSQLデータベース"],
next_steps: "プロジェクト構造を作成してから認証システムを実装"
});
技術仕様
- プロトコル: Model Context Protocol Elicitation
- 言語: TypeScript
- ランタイム: Node.js
- SDKバージョン: @modelcontextprotocol/sdk ^1.0.0
プロトコルの仕組み
このサーバーは真のMCP Elicitationプロトコルを実装しており：
elicitation/createメソッドでクライアントにリクエスト送信- JSONスキーマでユーザー入力の構造を定義
- ユーザーは
accept、decline、cancelで応答 acceptの場合、スキーマに従った構造化データを受信
これにより、AIとユーザー間の確実な意思疎通が可能になります。