MCP of MCPs

import { pipeline, type FeatureExtractionPipeline } from '@xenova/transformers'; export class EmbeddingsManager { private embedder: FeatureExtractionPipeline | undefined; private modelName: string = 'Xenova/all-MiniLM-L6-v2'; private initialized: boolean = false; constructor() {} /** * Initialize the embedding model */ async initialize(): Promise<void> { if (this.initialized) return; console.error('Initializing embedding model...'); // Load embedding model this.embedder = await pipeline( 'feature-extraction', this.modelName ); this.initialized = true; console.error('✓ Embedding model initialized'); } /** * Generate embedding for a text */ async embed(text: string): Promise<number[]> { if (!this.initialized || !this.embedder) { throw new Error('EmbeddingsManager not initialized. Call initialize() first.'); } // Generate embedding const output = await this.embedder(text, { pooling: 'mean', normalize: true, }); // Convert to array const embedding = Array.from(output.data as Float32Array); return embedding; } /** * Batch embed multiple texts */ async embedBatch(texts: string[]): Promise<number[][]> { const embeddings: number[][] = []; for (const text of texts) { const embedding = await this.embed(text); embeddings.push(embedding); } return embeddings; } }