Prompt Optimizer MCP Server

#!/usr/bin/env node /** * 提示词优化 MCP 服务器 * * 这个服务器展示了模型上下文协议 (MCP) 的核心概念： * 1. 服务器 (Server): 处理请求的主要 MCP 服务器实例 * 2. 工具 (Tools): 可被 AI 模型调用的函数 (optimize_prompt) * 3. 传输 (Transport): 通信层 (StdioServerTransport 用于命令行使用) * 4. 模式 (Schema): 使用 JSON Schema 定义工具的输入输出 */ import { Server } from "@modelcontextprotocol/sdk/server/index.js"; import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js"; import { CallToolRequestSchema, ListToolsRequestSchema, } from "@modelcontextprotocol/sdk/types.js"; // 类型定义，提供更好的 TypeScript 支持 interface OptimizationRequest { originalPrompt: string; // 原始提示词 context?: string; // 上下文信息 targetAudience?: string; // 目标受众 optimizationGoals?: OptimizationGoal[]; // 优化目标 style?: string; // 风格 } interface OptimizationResult { optimizedPrompt: string; // 优化后的提示词 improvements: string[]; // 改进列表 reasoning: string; // 推理过程 originalLength: number; // 原始长度 optimizedLength: number; // 优化后长度 score: number; // 评分 } interface PromptAnalysis { length: number; // 长度 hasContext: boolean; // 是否有上下文 hasExamples: boolean; // 是否有示例 hasConstraints: boolean; // 是否有约束 hasFormat: boolean; // 是否有格式要求 clarity: number; // 清晰度评分 specificity: number; // 具体性评分 } type OptimizationGoal = | 'clarity' // 清晰度 | 'specificity' // 具体性 | 'structure' // 结构 | 'context' // 上下文 | 'actionable' // 可执行性 | 'conciseness' // 简洁性 | 'examples' // 示例 | 'format'; // 格式 /** * PromptOptimizer 类 - 提示词优化的核心业务逻辑 * 这展示了 MCP 服务器中关注点分离的设计 */ class PromptOptimizer { private readonly optimizationStrategies: Record<OptimizationGoal, string> = { clarity: "使提示词更清晰和具体", specificity: "添加具体细节和约束条件", structure: "改善逻辑结构和流程", context: "提供更好的上下文和背景信息", actionable: "使指令更具可执行性和具体性", conciseness: "在保持清晰度的同时去除冗余", examples: "添加相关示例来说明期望", format: "明确指定所需的输出格式" }; async optimizePrompt(request: OptimizationRequest): Promise<OptimizationResult> { const { originalPrompt, context, targetAudience, optimizationGoals, style } = request; // 分析原始提示词 const analysis = this.analyzePrompt(originalPrompt); // 应用优化策略 let optimizedPrompt = originalPrompt; const improvements: string[] = []; // 应用选定的优化目标或默认目标 const goals = optimizationGoals || ['clarity', 'specificity', 'structure']; for (const goal of goals) { const result = this.applyOptimization(optimizedPrompt, goal, context, targetAudience, style); optimizedPrompt = result.prompt; if (result.improvement) { improvements.push(result.improvement); } } // 计算改进评分 const score = this.calculateScore(originalPrompt, optimizedPrompt, improvements); return { optimizedPrompt, improvements, reasoning: this.generateReasoning(originalPrompt, optimizedPrompt, improvements), originalLength: originalPrompt.length, optimizedLength: optimizedPrompt.length, score }; } private analyzePrompt(prompt: string): PromptAnalysis { return { length: prompt.length, hasContext: prompt.toLowerCase().includes('context') || prompt.toLowerCase().includes('background'), hasExamples: prompt.toLowerCase().includes('example') || prompt.toLowerCase().includes('for instance'), hasConstraints: prompt.toLowerCase().includes('must') || prompt.toLowerCase().includes('should') || prompt.toLowerCase().includes('requirement'), hasFormat: prompt.toLowerCase().includes('format') || prompt.toLowerCase().includes('structure'), clarity: this.assessClarity(prompt), specificity: this.assessSpecificity(prompt) }; } private assessClarity(prompt: string): number { // 清晰度评估的简单启发式方法 const vagueWords = ['thing', 'stuff', 'something', 'anything', 'maybe', 'perhaps', 'kind of', 'sort of']; const vagueCount = vagueWords.reduce((count, word) => count + (prompt.toLowerCase().split(word).length - 1), 0); return Math.max(0, 10 - vagueCount * 2); } private assessSpecificity(prompt: string): number { // 具体性评估的简单启发式方法 const specificWords = ['exactly', 'precisely', 'specifically', 'must', 'should', 'required', 'include', 'exclude']; const specificCount = specificWords.reduce((count, word) => count + (prompt.toLowerCase().split(word).length - 1), 0); return Math.min(10, specificCount); } private applyOptimization( prompt: string, goal: OptimizationGoal, context?: string, targetAudience?: string, style?: string ): { prompt: string; improvement?: string } { let optimizedPrompt = prompt; let improvement: string | undefined; switch (goal) { case 'clarity': if (this.assessClarity(prompt) < 7) { optimizedPrompt = this.improveClarity(prompt); improvement = "通过移除模糊语言和添加具体指令来增强清晰度"; } break; case 'specificity': if (this.assessSpecificity(prompt) < 5) { optimizedPrompt = this.improveSpecificity(prompt, context); improvement = "添加了具体的约束条件和要求"; } break; case 'structure': optimizedPrompt = this.improveStructure(prompt); improvement = "重新组织内容，使其具有清晰的章节和逻辑流程"; break; case 'context': if (context) { optimizedPrompt = this.addContext(prompt, context); improvement = "添加了相关的上下文和背景信息"; } break; case 'actionable': optimizedPrompt = this.makeActionable(prompt); improvement = "使指令更加具体和可执行"; break; case 'examples': optimizedPrompt = this.addExamples(prompt); improvement = "添加示例来澄清期望"; break; case 'format': optimizedPrompt = this.specifyFormat(prompt); improvement = "指定了清晰的输出格式要求"; break; } if (targetAudience) { optimizedPrompt = this.adaptForAudience(optimizedPrompt, targetAudience); } if (style) { optimizedPrompt = this.adaptStyle(optimizedPrompt, style); } return { prompt: optimizedPrompt, improvement }; } private improveClarity(prompt: string): string { // 用更具体的语言替换模糊术语 let improved = prompt .replace(/\bthing\b/gi, 'item') .replace(/\bstuff\b/gi, 'content') .replace(/\bsomething\b/gi, 'a specific item') .replace(/\bmaybe\b/gi, 'if applicable') .replace(/\bkind of\b/gi, 'similar to'); // 添加清晰度标记 if (!improved.includes('Please') && !improved.includes('I need')) { improved = 'Please ' + improved.charAt(0).toLowerCase() + improved.slice(1); } return improved; } private improveSpecificity(prompt: string, context?: string): string { let improved = prompt; // 如果缺少具体要求则添加 if (!improved.toLowerCase().includes('must') && !improved.toLowerCase().includes('should')) { improved += '\n\n要求:\n- 要具体和详细\n- 提供清晰的解释'; } if (context) { improved = `上下文: ${context}\n\n${improved}`; } return improved; } private improveStructure(prompt: string): string { // 如果提示词很长且无结构，则添加章节 if (prompt.length > 200 && !prompt.includes('\n\n')) { const sentences = prompt.split('. '); if (sentences.length > 3) { const intro = sentences.slice(0, 2).join('. ') + '.'; const details = sentences.slice(2).join('. '); return `${intro}\n\n详细信息:\n${details}`; } } return prompt; } private addContext(prompt: string, context: string): string { return `上下文: ${context}\n\n任务: ${prompt}`; } private makeActionable(prompt: string): string { let improved = prompt; // 如果缺少动作动词则添加 const actionVerbs = ['analyze', 'create', 'generate', 'explain', 'describe', 'list', 'compare']; const hasActionVerb = actionVerbs.some(verb => improved.toLowerCase().includes(verb)); if (!hasActionVerb) { improved = 'Please analyze and ' + improved.charAt(0).toLowerCase() + improved.slice(1); } return improved; } private addExamples(prompt: string): string { if (!prompt.toLowerCase().includes('example')) { return prompt + '\n\n请提供具体示例来说明你的回答。'; } return prompt; } private specifyFormat(prompt: string): string { if (!prompt.toLowerCase().includes('format') && !prompt.toLowerCase().includes('structure')) { return prompt + '\n\n格式: 请用标题和要点清晰地组织你的回答。'; } return prompt; } private adaptForAudience(prompt: string, audience: string): string { if (prompt.includes(`目标受众: ${audience}`)) { return prompt; // 已经适配 } const audienceNote = `\n\n目标受众: ${audience} - 请相应调整复杂度和术语。`; return prompt + audienceNote; } private adaptStyle(prompt: string, style: string): string { if (prompt.includes(`风格: ${style}`)) { return prompt; // 已经适配 } const styleNote = `\n\n风格: ${style}`; return prompt + styleNote; } private calculateScore(original: string, optimized: string, improvements: string[]): number { // 基于改进情况的简单评分 let score = 5; // 基础分数 score += improvements.length * 1.5; // 每个改进增加分数 // 长度优化奖励 if (optimized.length > original.length && optimized.length < original.length * 1.5) { score += 1; // 良好的扩展 } else if (optimized.length < original.length && optimized.length > original.length * 0.8) { score += 1; // 良好的压缩 } return Math.min(10, Math.round(score * 10) / 10); } private generateReasoning(original: string, optimized: string, improvements: string[]): string { let reasoning = "优化分析:\n\n"; reasoning += `原始提示词分析:\n`; reasoning += `- 长度: ${original.length} 个字符\n`; reasoning += `- 清晰度评分: ${this.assessClarity(original)}/10\n`; reasoning += `- 具体性评分: ${this.assessSpecificity(original)}/10\n\n`; reasoning += `应用的改进:\n`; improvements.forEach((improvement, index) => { reasoning += `${index + 1}. ${improvement}\n`; }); reasoning += `\n结果: 增强了提示词的结构、清晰度和可执行性。`; return reasoning; } } /** * MCP 工具定义 * 这定义了 optimize_prompt 工具的模式和元数据 * 展示的关键 MCP 概念: * - 工具名称和描述，供 AI 模型理解 * - 用于输入验证的 JSON Schema * - 必需参数与可选参数 */ const OPTIMIZE_PROMPT_TOOL = { name: "optimize_prompt", description: "优化和增强提示词以获得更好的 AI 交互效果。此工具分析提示词并应用各种优化策略来改善清晰度、具体性、结构和有效性。", inputSchema: { type: "object", properties: { originalPrompt: { type: "string", description: "要优化的原始提示词" }, context: { type: "string", description: "优化过程中要考虑的额外上下文或背景信息" }, targetAudience: { type: "string", description: "提示词的目标受众 (例如: '技术专家', '普通用户', '学生')" }, optimizationGoals: { type: "array", items: { type: "string", enum: ["clarity", "specificity", "structure", "context", "actionable", "conciseness", "examples", "format"] }, description: "要专注的具体优化目标。可用选项: clarity(清晰度), specificity(具体性), structure(结构), context(上下文), actionable(可执行性), conciseness(简洁性), examples(示例), format(格式)" }, style: { type: "string", description: "优化后提示词的期望风格 (例如: 'formal(正式)', 'casual(随意)', 'technical(技术性)', 'creative(创意性)')" } }, required: ["originalPrompt"] } }; /** * MCP 服务器设置 * 这展示了核心的 MCP 服务器架构: * 1. 带有元数据的服务器实例 * 2. 能力声明 * 3. 不同 MCP 操作的请求处理器 */ const server = new Server({ name: "prompt-optimizer-server", version: "0.1.0", }, { capabilities: { tools: {}, // 表示此服务器提供工具 }, }); // 创建提示词优化器实例 const promptOptimizer = new PromptOptimizer(); /** * 列出工具处理器 * MCP 服务器必须实现此功能来告诉客户端有哪些工具可用 */ server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools: [OPTIMIZE_PROMPT_TOOL], })); /** * 调用工具处理器 * 这处理来自 MCP 客户端的实际工具执行请求 * 展示错误处理和响应格式化 */ server.setRequestHandler(CallToolRequestSchema, async (request) => { if (request.params.name === "optimize_prompt") { try { const args = request.params.arguments; if (!args || typeof args !== 'object' || !('originalPrompt' in args)) { throw new Error('缺少必需参数: originalPrompt'); } const optimizationRequest: OptimizationRequest = { originalPrompt: args.originalPrompt as string, context: args.context as string | undefined, targetAudience: args.targetAudience as string | undefined, optimizationGoals: args.optimizationGoals as OptimizationGoal[] | undefined, style: args.style as string | undefined, }; const result = await promptOptimizer.optimizePrompt(optimizationRequest); return { content: [{ type: "text", text: JSON.stringify(result, null, 2) }] }; } catch (error) { return { content: [{ type: "text", text: `优化提示词时出错: ${error instanceof Error ? error.message : String(error)}` }], isError: true }; } } return { content: [{ type: "text", text: `未知工具: ${request.params.name}` }], isError: true }; }); /** * 服务器启动 * 这展示了使用 stdio 进行 CLI 使用的 MCP 传输设置 * 服务器通过 stdin/stdout 通信，使其与各种 MCP 客户端兼容， * 包括 Claude Desktop 和 Cursor */ async function runServer() { const transport = new StdioServerTransport(); await server.connect(transport); console.error("提示词优化 MCP 服务器正在 stdio 上运行"); } // 启动服务器 runServer().catch((error) => { console.error("运行服务器时发生致命错误:", error); process.exit(1); });