Lead Qualifier MCP Tool

import { chat } from "../llm.js"; import { PROMPTS } from "../constants.js"; // TODO: support sessions from Redis const sessions = new Map(); const weights = { budget: 30, authority: 20, need: 30, timeline: 20, }; const allFields = ["budget", "authority", "need", "timeline"]; // TODO: move this to Redis cache per tenant/language const leadQualificationQuestions = { budget: "What monthly budget range are you considering for a solution like this? (e.g., under $100, $100–$500, $500+)", authority: "Are you the main person evaluating tools like this, or is there someone else involved in the decision?", need: "What specific problems are you hoping to solve with a solution like this?", timeline: "When are you looking to implement a solution?" }; function getSession(sessionId) { if (!sessions.has(sessionId)) { sessions.set(sessionId, { qualificationMap: {}, scoreMap: {}, totalScore: 0, nextField: allFields[0], lastPromptedField: null }); } return sessions.get(sessionId); } function getMissingFields(session) { return allFields.filter((field) => !session.qualificationMap[field]); } function updateSession(sessionId, { field, value, score, nextField, lastPromptedQuestion }) { const session = getSession(sessionId); if (field) { const weight = weights[field] ?? 0; const weightedScore = Math.round(score * weight / 100); const prevScore = session.scoreMap[field] ?? 0; session.totalScore = (session.totalScore ?? 0) - prevScore + weightedScore; session.qualificationMap[field] = value; session.scoreMap[field] = weightedScore; } session.nextField = nextField; session.lastPromptedQuestion = lastPromptedQuestion; } function printSession(sessionId) { console.log("[%s]Session: %s", sessionId, JSON.stringify(getSession(sessionId), null, 2)); } /** * Main function to run the lead qualifier */ export async function runLeadQualifier(sessionId, message) { console.log("[%s]Running lead qualifier - %s", sessionId, message); const session = getSession(sessionId); // Step 1: Extract qualification info from message const prompt = PROMPTS.leadQualificationExtractor(allFields, session.lastPromptedQuestion, message); const response = await chat(prompt); let parsed; try { const cleaned = response.trim().replace(/^```(?:json)?\s*/i, "").replace(/```$/, ""); parsed = JSON.parse(cleaned); } catch (err) { console.error("[%s][LLM]Failed to parse LLM response:", sessionId, response); printSession(sessionId); return "Sorry, I couldn’t understand your response. Could you rephrase?" } const { field, value, score } = parsed; if (!field || !value || typeof score !== "number") { console.log("[%s][LLM]Cannot extract valid info from message: %s", sessionId, message); } else { console.log("[%s][LLM]Extracted info from message: %s", sessionId, JSON.stringify({ field, value, score }, null, 2)); } // Step 2: Get next follow-up question const missingFields = getMissingFields(session); const nextField = missingFields.filter((f) => f !== field)[0] || null; const nextFieldQuestion = nextField ? leadQualificationQuestions[nextField] : null; // Step 3: Update session updateSession(sessionId, { field, value, score, nextField, lastPromptedQuestion: nextFieldQuestion }); const result = nextFieldQuestion ? nextFieldQuestion : "Thanks! I’ve collected everything I need for now." printSession(sessionId); return result; } export function clearSession(sessionId) { sessions.delete(sessionId); printSession(sessionId); }