ailistmybusiness

MCP-callable directory for AI-driven SMB discovery. Country-agnostic, zero-PII catalog of realtors, insurance agents, and medical practitioners (Wave 1).

Working title. Public brand pending domain registration. Folder name and package.json will be renamed once the domain is locked.

What this is

When a user asks ChatGPT, Claude, or Gemini "find me a realtor in Dallas" / "evening walk-in clinic in Toronto" / "bilingual insurance broker", the AI calls this MCP server. It returns ranked business listings with UTM-tagged booking URLs. The user books directly with the SMB. We never see customer data — we are a business catalog, not a lead processor.

Phase 1 status

• MCP server scaffold (Node 20 + TypeScript)

• 5 tools: search_businesses, get_business_profile, get_booking_options, search_by_query, get_categories

• 30 mock SMBs across 3 verticals × 2 countries (Dallas + Toronto)

• OpenStreetMap Nominatim geocoding (free, no PII)

• UTM-tagged booking URLs for SMB attribution

• Vitest test suite for all 5 tools

• Smithery + Glama + Railway manifests

• Supabase wiring (Phase 2)

• Stripe billing for paid tiers (Phase 2)

• Coinbase x402 metering for API tier (Phase 3)

Quick start

npm install
npm test                    # run unit tests
npm run test:mcp            # smoke test all tools end-to-end
npm run dev                 # start MCP server on stdio
npm run http                # start HTTP server on :3000 (preview endpoints + Railway entrypoint)

Then visit http://localhost:3000/preview/search?category=realtor&location=Dallas to see the ranking output.

Architecture

src/
  server.ts              # MCP server (stdio transport, Smithery entrypoint)
  http.ts                # Express server (Railway entrypoint, /health, /preview/*)
  types.ts               # BusinessProfile, SearchHit, BookingOptions, etc.
  tools/                 # one file per MCP tool
    searchBusinesses.ts
    getBusinessProfile.ts
    getBookingOptions.ts
    searchByQuery.ts
    getCategories.ts
  lib/
    db.ts                # data access — switches on DATA_SOURCE env (mock | supabase)
    ranking.ts           # 6-factor weighted relevance score
    utm.ts               # UTM URL builder for booking links
    geo.ts               # OpenStreetMap Nominatim geocoder + Haversine distance
data/
  mockBusinesses.json    # 30 sample SMBs (realtors, insurance, medical × Dallas, Toronto)
  categories.json        # vertical taxonomy
scripts/
  seed.ts                # Supabase seeder (Phase 2 stub)
  test-mcp.ts            # smoke test runner
tests/
  tools.test.ts          # Vitest tests for all tools

Zero-PII rule

This catalog stores business data only:

• Business name, address, hours, services

• Public credentials and license numbers

• Aggregate review counts and ratings (sourced from public APIs in Phase 2)

• UTM-tagged booking URLs

It explicitly does not store:

• Customer / patient names, phones, emails, or any other identifiers

• Insurance policy details, medical history, or anything covered by HIPAA / PIPEDA / GDPR

• Individual booking records or appointment data

Booking flow: agent gets the SMB's booking URL → user clicks → user books on the SMB's own system. We see impressions; SMB sees conversions via UTM tags on their own analytics.

MCP tool contracts

search_businesses

{
  category: string,         // "realtor" | "insurance_agent" | "medical_practitioner" | etc.
  location: string,         // "Dallas, TX" — geocoded server-side
  countryCode?: "US" | "CA" | "GB" | "AU" | ...,
  language?: string,        // ISO-639-1, e.g. "en", "fr", "es"
  subcategory?: string,
  maxResults?: number,      // default 10, max 25
  minRating?: number
}
→ SearchHit[]

get_business_profile

{ id: string, agentName?: string }
→ BusinessProfile  // bookingUrl is UTM-tagged

get_booking_options

{ id: string, agentName?: string }
→ { bookingUrl, acceptedMethods, hours, timezone, fallbackContact }

search_by_query

{ query: string, location?: string, countryCode?: string, maxResults?: number }
→ SearchHit[]

Phase 1 implementation is keyword/substring-based. Phase 2 swaps in pgvector or OpenAI embeddings for true semantic search.

get_categories

{ countryCode?: string }
→ CategoryEntry[]

Ranking logic

Weighted score (0–100) per business:

• Tier (20%) — healthcare 100 / pro 85 / standard 65 / free 40

• Distance (30%) — closer to query origin scores higher

• Rating (20%) — public review rating × volume

• Vertical / subcategory match (20%)

• Verified listing (5%)

• Language match (5%)

See src/lib/ranking.ts.

Hand-off to Claude Code

Once you clone this folder into your local dev directory:

# 1. Install
npm install

# 2. Initialize git
git init
git add .
git commit -m "Initial scaffold: MCP server + 5 tools + mock data"
git branch -M main
git remote add origin git@github.com:YOUR_GH_USERNAME/ailistmybusiness.git
git push -u origin main

# 3. Validate locally
npm run typecheck
npm test
npm run test:mcp

# 4. Submit to Smithery (when ready)
#    https://smithery.ai/new — point to your GitHub repo

# 5. Deploy HTTP entrypoint to Railway (when ready)
#    https://railway.app/new — uses railway.json

Phase 2 plug-in points

When you're ready to wire real services:

License

Proprietary. © 2026 Charles Mutamiri.