Structured Workflow Engine MCP Server

import fs from 'fs'; import path from 'path'; import OpenAI from 'openai'; interface WorkflowEmbedding { id: string; title: string; description: string; category: string; tags: string[]; complexity: string; searchableText: string; embedding: number[]; } interface SearchResult { id: string; title: string; description: string; category: string; tags: string[]; complexity: string; similarity: number; } const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY }); // Always use public/workflow-embeddings.json for consistency const EMBEDDINGS_FILE = path.join(process.cwd(), 'public', 'workflow-embeddings.json'); export class SemanticSearch { private embeddings: WorkflowEmbedding[] = []; private loaded = false; /** * Load embeddings from cache file */ private async loadEmbeddings(): Promise<void> { if (this.loaded) return; console.log(`[SemanticSearch] Attempting to load embeddings from: ${EMBEDDINGS_FILE}`); if (!fs.existsSync(EMBEDDINGS_FILE)) { console.warn('[SemanticSearch] Embeddings file not found. Run "npm run build:embeddings" first.'); console.log(`[SemanticSearch] Current working directory: ${process.cwd()}`); console.log(`[SemanticSearch] Looking for file at: ${EMBEDDINGS_FILE}`); // Try alternative paths for production const altPaths = [ path.join(process.cwd(), 'workflow-embeddings.json'), path.join(process.cwd(), 'public', 'workflow-embeddings.json'), path.join(process.cwd(), '.next', 'static', 'workflow-embeddings.json') ]; for (const altPath of altPaths) { if (fs.existsSync(altPath)) { console.log(`[SemanticSearch] Found embeddings at alternative path: ${altPath}`); const embeddingsData = fs.readFileSync(altPath, 'utf-8'); this.embeddings = JSON.parse(embeddingsData); this.loaded = true; console.log(`[SemanticSearch] Loaded ${this.embeddings.length} workflow embeddings`); return; } } console.log(`[SemanticSearch] Checked alternative paths:`, altPaths); return; } try { const embeddingsData = fs.readFileSync(EMBEDDINGS_FILE, 'utf-8'); this.embeddings = JSON.parse(embeddingsData); this.loaded = true; console.log(`[SemanticSearch] Loaded ${this.embeddings.length} workflow embeddings`); } catch (error) { console.error('[SemanticSearch] Error loading embeddings:', error); } } /** * Search workflows using semantic similarity */ async searchWorkflows(query: string, limit: number = 5): Promise<SearchResult[]> { await this.loadEmbeddings(); if (this.embeddings.length === 0) { console.warn('[SemanticSearch] No embeddings available for search'); return []; } if (!process.env.OPENAI_API_KEY) { console.warn('[SemanticSearch] OpenAI API key not available, falling back to text search'); return this.fallbackTextSearch(query, limit); } try { // Generate embedding for search query console.log(`[SemanticSearch] Generating query embedding for: "${query}"`); const queryEmbeddingResponse = await openai.embeddings.create({ model: 'text-embedding-3-small', input: query.toLowerCase(), encoding_format: 'float' }); const queryEmbedding = queryEmbeddingResponse.data[0].embedding; // Calculate similarities with all workflow embeddings const similarities = this.embeddings.map(workflow => { const similarity = this.cosineSimilarity(queryEmbedding, workflow.embedding); return { id: workflow.id, title: workflow.title, description: workflow.description, category: workflow.category, tags: workflow.tags, complexity: workflow.complexity, similarity }; }); // Sort by similarity and return top results const results = similarities .sort((a, b) => b.similarity - a.similarity) .slice(0, limit); console.log(`[SemanticSearch] Found ${results.length} results with similarities:`, results.map(r => `${r.id}: ${Math.round(r.similarity * 100)}%`)); return results; } catch (error) { console.error('[SemanticSearch] Error in semantic search:', error); // Fallback to text search return this.fallbackTextSearch(query, limit); } } /** * Fallback text-based search when semantic search fails */ private fallbackTextSearch(query: string, limit: number): SearchResult[] { console.log('[SemanticSearch] Using fallback text search'); const searchTerms = query.toLowerCase().split(' '); const scored = this.embeddings.map(workflow => { let score = 0; const searchableText = `${workflow.title} ${workflow.description} ${workflow.category} ${workflow.tags.join(' ')}`.toLowerCase(); searchTerms.forEach(term => { if (searchableText.includes(term)) { score += 1; // Boost score for title matches if (workflow.title.toLowerCase().includes(term)) { score += 2; } // Boost score for category matches if (workflow.category.toLowerCase().includes(term)) { score += 1.5; } } }); // Normalize score to 0-1 range for consistency const normalizedSimilarity = Math.min(score / searchTerms.length, 1); return { id: workflow.id, title: workflow.title, description: workflow.description, category: workflow.category, tags: workflow.tags, complexity: workflow.complexity, similarity: normalizedSimilarity }; }); return scored .filter(item => item.similarity > 0) .sort((a, b) => b.similarity - a.similarity) .slice(0, limit); } /** * Calculate cosine similarity between two vectors */ private cosineSimilarity(vecA: number[], vecB: number[]): number { if (vecA.length !== vecB.length) { throw new Error('Vectors must have the same length'); } let dotProduct = 0; let normA = 0; let normB = 0; for (let i = 0; i < vecA.length; i++) { dotProduct += vecA[i] * vecB[i]; normA += vecA[i] * vecA[i]; normB += vecB[i] * vecB[i]; } normA = Math.sqrt(normA); normB = Math.sqrt(normB); if (normA === 0 || normB === 0) { return 0; } return dotProduct / (normA * normB); } /** * Get all available workflow categories */ async getAvailableCategories(): Promise<string[]> { await this.loadEmbeddings(); const categories = [...new Set(this.embeddings.map(w => w.category))]; return categories.sort(); } /** * Get workflows by category */ async getWorkflowsByCategory(category: string): Promise<SearchResult[]> { await this.loadEmbeddings(); return this.embeddings .filter(w => w.category.toLowerCase() === category.toLowerCase()) .map(w => ({ id: w.id, title: w.title, description: w.description, category: w.category, tags: w.tags, complexity: w.complexity, similarity: 1.0 // Perfect match for category filter })); } /** * Clear embeddings cache (useful for testing) */ clearCache(): void { this.embeddings = []; this.loaded = false; } } // Export singleton instance export const semanticSearch = new SemanticSearch();