docs-mcp-server

/** * Ensures embeddings are persisted into the documents_vec virtual table. * * This test is self-contained: * - Uses a temporary SQLite database (storePath points to a temp dir) * - Uses MSW to mock OpenAI embeddings so it does not require network access */ import { afterAll, beforeAll, describe, expect, it } from "vitest"; import { mkdtempSync, rmSync } from "node:fs"; import path from "node:path"; import { tmpdir } from "node:os"; import { config } from "dotenv"; import Database from "better-sqlite3"; import * as sqliteVec from "sqlite-vec"; import { ScrapeTool } from "../src/tools/ScrapeTool"; import { createLocalDocumentManagement } from "../src/store"; import { PipelineFactory } from "../src/pipeline/PipelineFactory"; import { EmbeddingConfig, type EmbeddingModelConfig, } from "../src/store/embeddings/EmbeddingConfig"; import { EventBusService } from "../src/events"; import { loadConfig } from "../src/utils/config"; config(); describe("Vector persistence", () => { let tempDir: string; let pipeline: any; let docService: any; let scrapeTool: ScrapeTool; const appConfig = loadConfig(); let prevOpenAiApiKey: string | undefined; let prevOpenAiApiBase: string | undefined; beforeAll(async () => { // Ensure vector search initializes in tests without requiring real credentials. prevOpenAiApiKey = process.env.OPENAI_API_KEY; prevOpenAiApiBase = process.env.OPENAI_API_BASE; process.env.OPENAI_API_KEY = process.env.OPENAI_API_KEY ?? "test-key"; delete process.env.OPENAI_API_BASE; tempDir = mkdtempSync(path.join(tmpdir(), "vector-persistence-e2e-")); const embeddingConfig: EmbeddingModelConfig = EmbeddingConfig.parseEmbeddingConfig( "openai:text-embedding-3-small", ); appConfig.app.storePath = tempDir; appConfig.app.embeddingModel = embeddingConfig.modelSpec; const eventBus = new EventBusService(); docService = await createLocalDocumentManagement(eventBus, appConfig); pipeline = await PipelineFactory.createPipeline(docService, eventBus, { appConfig, }); await pipeline.start(); scrapeTool = new ScrapeTool(pipeline); }, 30000); afterAll(async () => { if (pipeline) { await pipeline.stop(); } if (docService) { await docService.shutdown(); } if (tempDir) { try { rmSync(tempDir, { recursive: true, force: true }); } catch { // ignore cleanup errors } } if (prevOpenAiApiKey === undefined) { delete process.env.OPENAI_API_KEY; } else { process.env.OPENAI_API_KEY = prevOpenAiApiKey; } if (prevOpenAiApiBase === undefined) { delete process.env.OPENAI_API_BASE; } else { process.env.OPENAI_API_BASE = prevOpenAiApiBase; } }); it( "persists embeddings into documents_vec", async () => { const readmePath = path.resolve(process.cwd(), "README.md"); const fileUrl = `file://${readmePath}`; await scrapeTool.execute({ library: "vector-persist-lib", version: "1.0.0", url: fileUrl, waitForCompletion: true, }); const exists = await docService.exists("vector-persist-lib", "1.0.0"); expect(exists).toBe(true); const dbPath = path.join(tempDir, "documents.db"); const db = new Database(dbPath); sqliteVec.load(db); const { chunkCount } = db .prepare( "SELECT COUNT(*) as chunkCount FROM documents WHERE embedding IS NOT NULL", ) .get() as { chunkCount: number }; expect(chunkCount).toBeGreaterThan(0); const { vecCount } = db .prepare("SELECT COUNT(*) as vecCount FROM documents_vec") .get() as { vecCount: number }; expect(vecCount).toBeGreaterThan(0); expect(vecCount).toBe(chunkCount); }, 60000, ); });