Google Business Profile Review MCP Server

/** * LLM Service for generating review responses using MCP sampling */ import { logger } from '../utils/logger.js'; import { DEFAULTS } from '../utils/constants.js'; import { analyzeSentiment, extractEmotions, extractKeywords, calculateSentimentConfidence, calculateResponseConfidence, determineToneFromRating } from '../utils/sentimentAnalysis.js'; import { generateTemplateResponse, createReplyPrompt } from '../utils/templateGenerator.js'; import type { RequestHandlerExtra } from '@modelcontextprotocol/sdk/shared/protocol.js'; import type { GenerateReplyResponse, ServiceResponse, ReviewResponseContext } from '../types/index.js'; import { CreateMessageRequest, CreateMessageRequestSchema, ServerNotification, ServerRequest, ToolResultContent } from '@modelcontextprotocol/sdk/types.js'; type SendRequest = RequestHandlerExtra<ServerRequest, ServerNotification>['sendRequest']; const requestSampling = async (prompt: string, sendRequest: SendRequest) => { const request: CreateMessageRequest = { method: "sampling/createMessage", params: { messages: [ { role: "user", content: { type: "text", text: prompt } } ], systemPrompt: "Generate a professional and appropriate response to the customer review provided.", maxTokens: DEFAULTS.MAX_PROMPT_LENGTH, temperature: 0.7, includeContext: "thisServer" } }; const response: any = await sendRequest(request, CreateMessageRequestSchema); return response.content.text; }; export class LLMService { private samplingCallback?: (prompt: string) => Promise<string>; constructor(samplingCallback?: (prompt: string) => Promise<string>) { this.samplingCallback = samplingCallback; logger.debug('LLM Service initialized', { hasSampling: !!samplingCallback }); } /** * Set the sampling callback for LLM requests */ setSamplingCallback(callback: (prompt: string) => Promise<string>): void { this.samplingCallback = callback; logger.debug('Sampling callback registered'); } /** * Generate a reply to a review using LLM sampling * This method will use MCP's sampling capability to generate responses */ async generateReply( reviewText: string, starRating: number, businessName: string, options: { replyTone?: 'professional' | 'friendly' | 'apologetic' | 'grateful'; includePersonalization?: boolean; maxLength?: number; } = {}, extra: RequestHandlerExtra<ServerRequest, ServerNotification> ): Promise<ServiceResponse<GenerateReplyResponse>> { try { logger.debug('Generating reply for review', { starRating, businessName, hasSampling: !!this.samplingCallback }); const { replyTone = determineToneFromRating(starRating), includePersonalization = true, maxLength = DEFAULTS.MAX_PROMPT_LENGTH } = options; // Analyze sentiment const sentiment = analyzeSentiment(reviewText, starRating); // Create prompt for LLM const prompt = createReplyPrompt(reviewText, starRating, businessName, replyTone, includePersonalization); let replyText: string; let confidence: number; // Try to use LLM sampling if available if (this.samplingCallback) { try { logger.info('Using AI sampling for reply generation'); replyText = await this.samplingCallback(prompt); confidence = 0.9; // High confidence for AI-generated replies logger.info('AI-generated reply received', { length: replyText.length }); } catch (samplingError) { logger.warn('AI sampling failed, falling back to template', { error: samplingError }); replyText = generateTemplateResponse(reviewText, starRating, businessName, replyTone); confidence = calculateResponseConfidence(reviewText, starRating, replyText); } } else { logger.info('No AI sampling available, using template response'); replyText = await requestSampling(prompt, extra.sendRequest); confidence = 0.9; // High confidence for AI-generated replies } logger.info('Reply generated successfully', { sentiment, confidence, method: this.samplingCallback ? 'AI' : 'template' }); return { success: true, data: { replyText: replyText.substring(0, maxLength), tone: replyTone, sentiment, confidence } }; } catch (error) { logger.error('Error generating reply:', error); return { success: false, error: 'Failed to generate reply', errorCode: 'REPLY_GENERATION_ERROR' }; } } /** * Analyze review sentiment with detailed breakdown */ async analyzeSentimentDetailed(reviewText: string): Promise<ServiceResponse<{ sentiment: 'positive' | 'negative' | 'neutral'; confidence: number; emotions: string[]; keywords: string[]; }>> { try { logger.debug('Performing detailed sentiment analysis'); const sentiment = analyzeSentiment(reviewText, 3); // Neutral rating for text-only analysis const emotions = extractEmotions(reviewText); const keywords = extractKeywords(reviewText); const confidence = calculateSentimentConfidence(reviewText, sentiment); return { success: true, data: { sentiment, confidence, emotions, keywords } }; } catch (error) { logger.error('Error in sentiment analysis:', error); return { success: false, error: 'Failed to analyze sentiment', errorCode: 'SENTIMENT_ANALYSIS_ERROR' }; } } }