Google Business Profile Review MCP Server

# AI-Powered Reply Generation This MCP server now supports **two methods** for generating review replies: ## Method 1: Template-Based Replies (Current Default) The `generate_reply` tool uses pre-written templates: ``` Location: Noodle House Review: "It was awesome!" (5 stars) → Reply: "Thank you so much for your wonderful 5-star review! We're thrilled..." ``` **How it works:** - Automatic tone selection based on star rating - Pre-written professional templates - No AI API calls needed - Fast and reliable ## Method 2: AI-Generated Replies (Via MCP Prompts) Use the **`review_response` prompt** to generate AI-powered replies: ### In VS Code with GitHub Copilot: 1. **Access the prompt:** ``` @workspace /prompt review_response ``` 2. **Provide context:** - reviewText: "Pad thai is a must try, tastes very authentic..." - starRating: "5" - businessName: "Noodle House" - replyTone: "grateful" 3. **Copilot generates a personalized reply** using AI ### Why use AI-generated replies? - ✅ **Personalized** - References specific menu items, comments - ✅ **Natural** - Sounds more human and less templated - ✅ **Contextual** - Adapts to review sentiment and tone - ✅ **Constructive** - Addresses concerns in negative reviews ## How to Use Both Together ```bash # 1. Fetch reviews without replies get_reviews for locations/YOUR_LOCATION_ID # 2a. Quick template reply (fast) generate_reply + post_reply # 2b. AI-powered reply (better quality) Use @workspace /prompt review_response Then post_reply with the AI-generated text ``` ## Current Implementation Status ### ✅ Working Now: - Template-based reply generation - Review fetching (filters out replied reviews) - Posting replies to Google Business Profile - MCP prompts for AI assistance ### 🔄 Future Enhancement: - Direct integration with LLM APIs (OpenAI, Anthropic) - Automatic AI reply generation in the tool - Batch AI reply generation ## Configuration The LLM service is initialized in `src/server/mcpServer.ts`: ```typescript this.llmService = new LLMService(); ``` To add direct AI integration, you would: 1. Pass a sampling callback to `LLMService` 2. Implement the callback to call your chosen LLM API 3. The service will automatically use AI when available, falling back to templates ## Example Workflow 1. **Fetch unreplied reviews:** ```typescript mcp_google-busine_get_reviews(locationName) ``` 2. **Generate reply (template-based):** ```typescript mcp_google-busine_generate_reply({ reviewText: "It was awesome!", starRating: 5, businessName: "Noodle House" }) ``` 3. **Post the reply:** ```typescript mcp_google-busine_post_reply({ locationName: "locations/...", reviewId: "AbFvOq...", replyText: "Generated reply text" }) ``` ## Best Practices ### For Positive Reviews (4-5 stars): - Use "grateful" or "friendly" tone - Thank the customer - Reference specific items they mentioned - Invite them back ### For Negative Reviews (1-2 stars): - Use "apologetic" tone - Acknowledge their concerns - Offer to resolve offline - Provide contact information ### For Neutral Reviews (3 stars): - Use "professional" tone - Acknowledge feedback - Commit to improvement - Invite them to give another chance ## Reply Quality Metrics The system calculates a confidence score (0-1) based on: - Review length (more context = higher confidence) - Rating clarity (1 or 5 stars = higher confidence) - Reply appropriateness (length, tone match) Template-based: 0.7-0.9 confidence AI-generated: 0.9+ confidence (when available)