DEV Community

Luke | Fluke Studio
Luke | Fluke Studio

Posted on

I Built a UK Business Intelligence MCP Server — Here's How

#mcp #api #typescript #nextjs

Last month I needed UK business data for a project I was building. Company records, Google ratings, whether their website was actually live, social media links — the full picture.

The options were grim. Wire up 4 separate APIs yourself (Companies House, Google Places, DNS lookups, web scraping), each with different auth methods, rate limits, and response formats. Or pay an enterprise vendor like Creditsafe or OpenCorporates — starting at £2,250/year for API access.

I wanted something in between. A single API call that returns everything, at a price a solo dev can afford.

So I built it.

What It Does

Send a business name and location. Get back a complete profile from 4 data sources:

curl -X POST https://ukbusinessintel.com/api/v1/enrich \
  -H "X-API-Key: ukb_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"business_name": "Greggs", "location": "Newcastle"}'
Enter fullscreen mode Exit fullscreen mode

The response combines data from Companies House, Google Places, DNS/SSL checks, and social media scraping into a single JSON object:

{
  "companies_house": {
    "company_number": "01752681",
    "company_name": "GREGGS PLC",
    "company_status": "active",
    "incorporation_date": "1983-09-21",
    "sic_codes": ["10710", "47240"],
    "directors": [
      { "name": "SHERWOOD, Matthew James", "role": "director" },
      { "name": "SHERWOOD, Raymond Thomas", "role": "director" }
    ]
  },
  "google_places": {
    "display_name": "Greggs",
    "rating": 4.1,
    "review_count": 1847,
    "phone": "+44 191 281 7721",
    "website": "https://www.greggs.co.uk/"
  },
  "website": {
    "domain": "www.greggs.co.uk",
    "is_live": true,
    "http_status": 200,
    "ssl_valid": true,
    "ssl_expiry": "2026-04-15T23:59:59.000Z"
  },
  "social": {
    "facebook": "https://www.facebook.com/GreggsBakery",
    "instagram": "https://www.instagram.com/greaborggs_official",
    "linkedin": "https://www.linkedin.com/company/greggs",
    "twitter": "https://twitter.com/GreggsOfficial"
  },
  "meta": {
    "sources_successful": ["companies_house", "google_places", "dns", "scrape"],
    "duration_ms": 847
  }
}
Enter fullscreen mode Exit fullscreen mode

One call. Four sources. Under 2 seconds.

Architecture

The enrichment runs in two rounds of parallel execution:

Round 1 (parallel):
  ├── Companies House API → company records, directors, SIC codes
  └── Google Places API   → ratings, reviews, phone, website

Round 2 (parallel, only if domain discovered):
  ├── DNS/SSL Check       → is the site live? cert valid?
  └── Web Scraper         → social media links from homepage
Enter fullscreen mode Exit fullscreen mode

Round 2 depends on Round 1 because the business's website domain often comes from the Google Places result. If the user already provides a domain, both rounds run immediately.

Here's the core orchestration:

// Round 1: Companies House + Google Places in parallel
const [chResult, gpResult] = await Promise.allSettled([
  cachedFetch(companiesHouseAdapter, input),
  cachedFetch(googlePlacesAdapter, input),
]);

// Discover domain from Google Places website
if (!input.domain && gp.website) {
  input.domain = extractDomain(gp.website);
}

// Round 2: DNS + Web Scraper (only if domain available)
if (input.domain) {
  const [dnsResult, scrapeResult] = await Promise.allSettled([
    cachedFetch(dnsCheckAdapter, input),
    cachedFetch(webScraperAdapter, input),
  ]);
}
Enter fullscreen mode Exit fullscreen mode

Every adapter returns a consistent shape with success, data, and cached fields. If a source fails, the response still returns data from the sources that succeeded — you get partial results instead of an error.

Caching Without Redis

I didn't want to add Redis to the stack. Instead, all caching goes through Supabase (Postgres) with different TTLs per source:

  • Companies House: 7 days (company data changes slowly)
  • Google Places: 24 hours (ratings and reviews change more often)
  • DNS/SSL: 24 hours
  • Web scraping: 24 hours

Rate limiting also uses Postgres — atomic RPC functions handle the fixed-window counters instead of Redis INCR. One fewer service to manage.

The MCP Angle

This is also published as an MCP server on npm. If you use Claude Desktop or Cursor, you can add it to your config and ask Claude to look up any UK business:

{
  "mcpServers": {
    "uk-business-intelligence": {
      "command": "npx",
      "args": ["-y", "uk-business-intelligence-mcp"],
      "env": {
        "SUPABASE_URL": "your-url",
        "SUPABASE_SERVICE_ROLE_KEY": "your-key",
        "COMPANIES_HOUSE_API_KEY": "your-ch-key",
        "GOOGLE_PLACES_API_KEY": "your-google-key"
      }
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Then you can say "Tell me about Greggs in Newcastle" and Claude calls the tool, gets the enriched profile, and summarises it conversationally.

The MCP server exposes a single tool — enrich_uk_business — and calls the same enrichment orchestrator as the REST API. No HTTP round-trip, it runs the adapters directly.

Why MCP matters: there are a few Companies House MCP servers already out there, but they're thin wrappers around the Companies House API alone. This one combines 4 sources, handles caching, and degrades gracefully. That's the difference between "here's a company number lookup" and "here's everything you need to know about this business."

What I Learned

Google Places API pricing is a trap. The new Places API (post-2025) has tiered pricing. Basic fields like name and address are cheap. But the fields you actually want — ratings, reviews, phone number, website — are in the "Pro" tier at $20/1,000 requests. Plan your field masks carefully or you'll burn through budget fast.

Companies House API is free but quirky. The search endpoint returns fuzzy matches, so "Greggs" might return "Greggs PLC", "Greggs Properties Ltd", and "Mrs Greggs Bakery Ltd." I sort results by active companies first and match by name similarity.

Web scraping for social links is fragile. Every site structures their footer differently. I look for href patterns matching known social media domains (facebook.com, instagram.com, etc.) in the homepage HTML. It works for ~80% of businesses, which is good enough when it's one of four data sources.

You don't need Redis for rate limiting. A Postgres function with INSERT ... ON CONFLICT DO UPDATE and a timestamp check handles fixed-window rate limiting atomically. One fewer dependency, and it's fast enough for the throughput I need.

Try It

Free tier: 100 lookups/month, no credit card required.

If you work with UK business data — lead gen, CRM enrichment, due diligence, local SEO — I'd love to hear what data points you'd want added next. Open an issue or just try the free tier and let me know.

Top comments (0)