Local RAG

// RAG MCP Server E2E Test - Design Doc: rag-mcp-server-design.md (v1.1) // Generated: 2025-10-31 // Test Type: End-to-End Test // Implementation Timing: After all implementations complete import { mkdirSync, rmSync, writeFileSync } from 'node:fs' import { resolve } from 'node:path' import { describe, expect, it } from 'vitest' import { RAGServer } from '../../server/index.js' // ============================================ // E2E Test: Complete RAG Workflow // ============================================ describe('RAG MCP Server E2E Test', () => { // Note: Each test case uses independent RAGServer instance (to prevent inter-test interference) // -------------------------------------------- // -------------------------------------------- describe('Complete RAG Flow (Phase 1: PDF/TXT/MD)', () => { // User Scenario: File ingestion from MCP client → search workflow // Validation: End-to-end document ingestion → search → result retrieval works correctly it('User Journey: Ingest PDF file from MCP client → Search with natural language query → Retrieve related documents', async () => { // 1. Start MCP server (test case dedicated) const testDbPath = resolve('./tmp/e2e-lancedb-test1') const testDataDir = resolve('./tmp/e2e-data-test1') mkdirSync(testDbPath, { recursive: true }) mkdirSync(testDataDir, { recursive: true }) const ragServer = new RAGServer({ dbPath: testDbPath, modelName: 'Xenova/all-MiniLM-L6-v2', cacheDir: './tmp/models', baseDir: testDataDir, maxFileSize: 100 * 1024 * 1024, chunkSize: 512, chunkOverlap: 100, }) await ragServer.initialize() try { expect(ragServer).toBeDefined() // 2. Ingest TXT file (using TXT for testing since PDF parser requires actual PDF) const txtFile = resolve(testDataDir, 'sample-pdf.txt') writeFileSync( txtFile, 'This is a sample document for E2E testing. TypeScript type safety is important. ' + 'TypeScript provides better tooling at any scale. ' + 'TypeScript is a strongly typed programming language.' ) const ingestResult = await ragServer.handleIngestFile({ filePath: txtFile }) expect(ingestResult.content[0].text).toBeDefined() const ingestData = JSON.parse(ingestResult.content[0].text) expect(ingestData.chunkCount).toBeGreaterThan(0) expect(ingestData.filePath).toBe(txtFile) // 3. Search with natural language query const queryResult = await ragServer.handleQueryDocuments({ query: 'TypeScript type safety', limit: 5, }) const results = JSON.parse(queryResult.content[0].text) // 4. Verify related document retrieval expect(results.length).toBeGreaterThan(0) expect(results[0]).toHaveProperty('filePath') expect(results[0]).toHaveProperty('text') expect(results[0]).toHaveProperty('score') expect(results[0]).toHaveProperty('chunkIndex') // 5. Verify score ordering (ascending, LanceDB uses distance scores) for (let i = 0; i < results.length - 1; i++) { expect(results[i].score).toBeLessThanOrEqual(results[i + 1].score) } } finally { // Cleanup rmSync(testDbPath, { recursive: true, force: true }) rmSync(testDataDir, { recursive: true, force: true }) } }) }) // -------------------------------------------- // -------------------------------------------- describe('Complete RAG Flow (Phase 2: DOCX support, File re-ingestion)', () => { // User Scenario: Ingest DOCX file → Search // Validation: DOCX file can be ingested and searched it('User Journey: Ingest DOCX file → Search with natural language query → Retrieve related documents', async () => { // 1. Start MCP server (test case dedicated) const testDbPath = resolve('./tmp/e2e-lancedb-test2') const testDataDir = resolve('./tmp/e2e-data-test2') mkdirSync(testDbPath, { recursive: true }) mkdirSync(testDataDir, { recursive: true }) const ragServer = new RAGServer({ dbPath: testDbPath, modelName: 'Xenova/all-MiniLM-L6-v2', cacheDir: './tmp/models', baseDir: testDataDir, maxFileSize: 100 * 1024 * 1024, chunkSize: 512, chunkOverlap: 100, }) await ragServer.initialize() try { expect(ragServer).toBeDefined() // 2. Ingest DOCX file // Note: Create simple DOCX file (simulate with text instead of actual binary DOCX) const docxFile = resolve(testDataDir, 'sample.docx') writeFileSync( docxFile, 'This is a sample DOCX document for E2E testing. Project management is important. Agile methodology provides effective project management.' ) try { const ingestResult = await ragServer.handleIngestFile({ filePath: docxFile }) expect(ingestResult.content[0].text).toBeDefined() const ingestData = JSON.parse(ingestResult.content[0].text) expect(ingestData.chunkCount).toBeGreaterThan(0) // 3. Search with natural language query const queryResult = await ragServer.handleQueryDocuments({ query: 'project management', limit: 5, }) const results = JSON.parse(queryResult.content[0].text) // 4. Verify related document retrieval expect(results.length).toBeGreaterThan(0) expect(results[0].filePath).toContain('sample.docx') } catch (error) { // If DOCX parsing fails (when mammoth requires actual DOCX binary) // Verify error message expect((error as Error).message).toContain('Failed to parse DOCX') } } finally { // Cleanup rmSync(testDbPath, { recursive: true, force: true }) rmSync(testDataDir, { recursive: true, force: true }) } }) // User Scenario: File re-ingestion → Search // Validation: Update existing file and re-ingest, search with new content it('User Journey: Update existing file and re-ingest → Search with new content → Old content not included in search results', async () => { // 1. Start MCP server (test case dedicated) const testDbPath = resolve('./tmp/e2e-lancedb-test3') const testDataDir = resolve('./tmp/e2e-data-test3') mkdirSync(testDbPath, { recursive: true }) mkdirSync(testDataDir, { recursive: true }) const ragServer = new RAGServer({ dbPath: testDbPath, modelName: 'Xenova/all-MiniLM-L6-v2', cacheDir: './tmp/models', baseDir: testDataDir, maxFileSize: 100 * 1024 * 1024, chunkSize: 512, chunkOverlap: 100, }) await ragServer.initialize() try { expect(ragServer).toBeDefined() // 2. Initial file ingestion (old content: "TypeScript") const v1File = resolve(testDataDir, 'sample-v1.txt') writeFileSync( v1File, 'TypeScript is a strongly typed programming language that builds on JavaScript. It provides better tooling at any scale.' ) await ragServer.handleIngestFile({ filePath: v1File }) // 3. Verify search with old content const queryResult1 = await ragServer.handleQueryDocuments({ query: 'TypeScript', limit: 5, }) const results1 = JSON.parse(queryResult1.content[0].text) expect(results1.length).toBeGreaterThan(0) // 4. Update file (new content: "JavaScript") writeFileSync( v1File, 'JavaScript is a versatile programming language for web development. JavaScript provides dynamic features and flexibility.' ) // 5. Re-ingest file await ragServer.handleIngestFile({ filePath: v1File }) // 6. Verify search with new content const queryResult2 = await ragServer.handleQueryDocuments({ query: 'JavaScript', limit: 5, }) const results2 = JSON.parse(queryResult2.content[0].text) expect(results2.length).toBeGreaterThan(0) // 7. Verify old content not included in search results // Check file list to confirm sample-v1.txt exists only once (no duplicates) const listResult = await ragServer.handleListFiles() const files = JSON.parse(listResult.content[0].text) const targetFiles = files.filter((f: { filePath: string }) => f.filePath === v1File) // Validation: File exists only once (no duplicates) expect(targetFiles.length).toBe(1) // Validation: Chunk count exists (chunks after re-ingestion) expect(targetFiles[0].chunkCount).toBeGreaterThan(0) // 8. Search with new content and verify old content is not included const queryResult3 = await ragServer.handleQueryDocuments({ query: 'JavaScript versatile programming', limit: 5, }) const results3 = JSON.parse(queryResult3.content[0].text) // Verify sample-v1.txt is included in search results for new content "JavaScript is a versatile" expect(results3.length).toBeGreaterThan(0) const targetResult = results3.find((r: { filePath: string }) => r.filePath === v1File) expect(targetResult).toBeDefined() // Verify new content "JavaScript is a versatile" is included expect(targetResult.text).toContain('JavaScript') // Verify old content "TypeScript is a strongly typed" is not included expect(targetResult.text).not.toContain('TypeScript is a strongly typed') } finally { // Cleanup rmSync(testDbPath, { recursive: true, force: true }) rmSync(testDataDir, { recursive: true, force: true }) } }) }) })