AI MCP Gateway

/** * Web Researcher Agent * Performs web searches, extracts content from URLs, * and synthesizes research findings. * * Features: * - URL content extraction * - Search query generation * - Fact checking and source validation * - Research report generation * - n8n webhook compatible */ import type { AgentLLM } from './react.js'; import { logger } from '../logging/logger.js'; export interface WebSource { url: string; title?: string; content?: string; snippet?: string; fetchedAt?: number; } export interface ResearchQuery { topic: string; context?: string; maxSources?: number; focusAreas?: string[]; } export interface ResearchResult { success: boolean; topic: string; summary: string; keyFindings: string[]; sources: WebSource[]; confidence: number; processingTime: number; suggestedQueries?: string[]; } export interface FactCheckResult { claim: string; verdict: 'true' | 'false' | 'partially_true' | 'unverified'; explanation: string; sources: string[]; confidence: number; } /** * Web Researcher Agent * Conducts web research and synthesizes findings */ export class WebResearcher { private llm: AgentLLM; constructor(llm: AgentLLM) { this.llm = llm; } /** * Generate search queries for a research topic */ async generateSearchQueries(topic: string, count = 3): Promise<string[]> { const response = await this.llm.chat({ messages: [ { role: 'system', content: 'You are a research assistant. Generate effective search queries for the given topic. Return as JSON array of strings.', }, { role: 'user', content: `Topic: ${topic}\n\nGenerate ${count} search queries that would help find comprehensive information about this topic. Return JSON array only.`, }, ], maxTokens: 300, temperature: 0.3, }); try { const queries = JSON.parse(response.content); return Array.isArray(queries) ? queries : [topic]; } catch { return [topic]; } } /** * Extract and summarize content from provided URLs/sources */ async analyzeWebSources(sources: WebSource[]): Promise<{ summaries: Array<{ url: string; title: string; summary: string; keyPoints: string[] }>; overallThemes: string[]; }> { const summaries: Array<{ url: string; title: string; summary: string; keyPoints: string[] }> = []; const allKeyPoints: string[] = []; for (const source of sources.slice(0, 10)) { if (!source.content && !source.snippet) continue; const content = (source.content || source.snippet || '').slice(0, 3000); const response = await this.llm.chat({ messages: [ { role: 'system', content: 'Analyze the web content and extract key information. Return JSON: {"summary": "brief summary", "keyPoints": ["point1", "point2"]}', }, { role: 'user', content: `URL: ${source.url}\nTitle: ${source.title || 'Unknown'}\n\nContent:\n${content}`, }, ], maxTokens: 500, temperature: 0.1, }); try { const parsed = JSON.parse(response.content); summaries.push({ url: source.url, title: source.title || source.url, summary: parsed.summary || '', keyPoints: parsed.keyPoints || [], }); allKeyPoints.push(...(parsed.keyPoints || [])); } catch { summaries.push({ url: source.url, title: source.title || source.url, summary: response.content.slice(0, 200), keyPoints: [], }); } } // Extract overall themes const themesResponse = await this.llm.chat({ messages: [ { role: 'system', content: 'Identify common themes from these key points. Return JSON array of theme strings.', }, { role: 'user', content: `Key points from multiple sources:\n${allKeyPoints.join('\n')}\n\nIdentify 3-5 main themes.`, }, ], maxTokens: 200, temperature: 0.2, }); let overallThemes: string[] = []; try { overallThemes = JSON.parse(themesResponse.content); } catch { overallThemes = []; } return { summaries, overallThemes }; } /** * Conduct research on a topic using provided sources */ async research(query: ResearchQuery, providedSources: WebSource[] = []): Promise<ResearchResult> { const startTime = Date.now(); logger.info('Starting web research', { topic: query.topic }); // Analyze provided sources const analysis = providedSources.length > 0 ? await this.analyzeWebSources(providedSources) : { summaries: [], overallThemes: [] }; // Generate research summary const summaryPrompt = `Research Topic: ${query.topic} ${query.context ? `Context: ${query.context}` : ''} ${query.focusAreas ? `Focus Areas: ${query.focusAreas.join(', ')}` : ''} Source Summaries: ${analysis.summaries.map((s) => `- ${s.title}: ${s.summary}`).join('\n')} Key Points Found: ${analysis.summaries.flatMap((s) => s.keyPoints).join('\n')} Themes: ${analysis.overallThemes.join(', ')} Task: Provide a comprehensive research summary with key findings.`; const response = await this.llm.chat({ messages: [ { role: 'system', content: `You are a research analyst. Synthesize the information into a clear research report. Return JSON: {"summary": "comprehensive summary", "keyFindings": ["finding1", "finding2"], "confidence": 0.0-1.0, "suggestedQueries": ["query for more info"]}`, }, { role: 'user', content: summaryPrompt }, ], maxTokens: 1500, temperature: 0.2, }); try { const parsed = JSON.parse(response.content); return { success: true, topic: query.topic, summary: parsed.summary || '', keyFindings: parsed.keyFindings || [], sources: providedSources, confidence: parsed.confidence || 0.5, processingTime: Date.now() - startTime, suggestedQueries: parsed.suggestedQueries || [], }; } catch { return { success: true, topic: query.topic, summary: response.content, keyFindings: analysis.overallThemes, sources: providedSources, confidence: 0.5, processingTime: Date.now() - startTime, }; } } /** * Fact-check a claim against provided sources */ async factCheck(claim: string, sources: WebSource[]): Promise<FactCheckResult> { const sourceContent = sources .slice(0, 5) .map((s) => `Source: ${s.url}\n${(s.content || s.snippet || '').slice(0, 1500)}`) .join('\n\n---\n\n'); const response = await this.llm.chat({ messages: [ { role: 'system', content: `You are a fact-checker. Verify the claim against provided sources. Return JSON: {"verdict": "true|false|partially_true|unverified", "explanation": "why", "confidence": 0.0-1.0}`, }, { role: 'user', content: `Claim to verify: "${claim}"\n\nSources:\n${sourceContent}`, }, ], maxTokens: 500, temperature: 0.1, }); try { const parsed = JSON.parse(response.content); return { claim, verdict: parsed.verdict || 'unverified', explanation: parsed.explanation || '', sources: sources.map((s) => s.url), confidence: parsed.confidence || 0.5, }; } catch { return { claim, verdict: 'unverified', explanation: response.content, sources: sources.map((s) => s.url), confidence: 0.3, }; } } } export function createWebResearcher(llm: AgentLLM) { return new WebResearcher(llm); }