KiasuScout

KiasuScout is an MVP for Singapore and Southeast Asian businesses that want to know where parents' AI assistants send them.

It combines:

1. a Model Context Protocol server for agent workflows, and
2. a lightweight web frontend for running first-pass AI answer visibility reports.

The initial focus is middle-class parents looking for:

• tuition and academic support
• enrichment classes
• children's activities
• educational toys and learning products
• camps, workshops, STEM/arts/sports programmes

KiasuScout helps agencies and operators answer:

"When parents ask ChatGPT, Gemini, Perplexity or Google AI for recommendations, do we appear — or do our competitors?"

This MVP is intentionally measurement-first. It does not scrape consumer AI platforms yet. Instead, it provides prompt packs and analysis tools for answers captured manually, via approved APIs, or by later browser automation.

What's in the MVP

Parent-facing discovery loop

• Parent Scout flow for natural-language discovery questions
• Child age, location, category, budget and learning-goal context
• Seeded recommendation cards for Singapore tuition/enrichment/activity/toy providers
• Feedback capture: saved, contacted, too expensive, too far, not enough info, not suitable for age
• Local browser storage for demo feedback signals

Business-facing AEO intelligence

• prompt-pack generator for parent discovery queries
• form for business/category/location/competitors
• captured-answer JSON input
• answer-share report
• competitor mentions
• parent intent and objection summaries
• combined recommendations that translate parent feedback into AEO actions
• raw report JSON for export/debugging

Deployable web MVP

• FastAPI app for local/API-backed demos
• static Vercel build in public/index.html with in-browser fallback logic
• vercel.json for Vercel static deployment

MCP tools

• generate_prompt_pack — create Singapore/SEA parent-oriented prompt sets for a category/location.
• analyze_answer_visibility — parse captured LLM answers and score business visibility against competitors.
• recommend_visibility_fixes — produce practical local SEO/GEO fixes for the education/children sector.
• create_visibility_report — generate a complete client-ready JSON report.
• list_supported_segments — list supported locations, categories, parent personas, and platforms.

Install

git clone https://github.com/sixirixis/kiasu-scout.git
cd kiasu-scout
python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'

Run the web MVP locally

python -m answerspot_sg_mcp.web

Open:

http://127.0.0.1:8000

Deploy to Vercel

The repo includes a static Vercel entrypoint at public/index.html. The deployed demo keeps working even without the Python API because the browser has local fallback logic for parent search, prompt generation and reports.

npx vercel --prod

Run as an MCP server

python -m answerspot_sg_mcp.server

Example Hermes config:

mcp_servers:
  kiasu_scout:
    command: "python"
    args: ["-m", "answerspot_sg_mcp.server"]
    timeout: 120

Run tests

pytest -q
ruff check .

Example analysis payload

{
  "business_name": "Little Explorers STEM Club",
  "category": "STEM enrichment class",
  "location": "Tampines, Singapore",
  "competitors": ["The Learning Lab", "Saturday Kids", "Nullspace Robotics"],
  "answers": [
    {
      "platform": "ChatGPT",
      "prompt": "What are the best STEM enrichment classes in Tampines for a primary school child?",
      "answer_text": "Parents often compare Saturday Kids, Nullspace Robotics and Little Explorers STEM Club..."
    }
  ]
}

Product direction

The initial ICP is local SEO agencies and education/enrichment operators in Singapore. The first commercial deliverable should be a white-label monthly AI visibility report showing:

• answer share across platforms
• competitor recommendations
• prompt/category gaps
• cited sources and reputation signals
• recommended fixes: Google Business Profile, local directories, parent forums, review targets, schema, service pages, FAQs, and marketplace listings

Safety and terms

This repo does not include scraping logic. Any future connector should prefer official APIs or user-authorized collection and clearly disclose methodology.