parkpilot-mcp

A small remote MCP server that exposes one tool, carpark_availability, backed by LTA DataMall's live CarParkAvailabilityv2 feed. Built so the ParkPilot artifact can call it through claude.ai's mcp capability and show real available-lot counts instead of sample data.

Coverage note: LTA DataMall's carpark feed mainly covers HDB multi-storey/surface carparks and some LTA/URA lots. It does not include most private mall or office-building carparks (Suntec, Wisma Atria, Marina Bay Financial Centre, etc.) — there is no free public live-availability feed for those. So "real data" here means real HDB carpark availability near a destination, not the specific malls used in the demo.

1. Get an LTA DataMall API key

1. Go to https://datamall.lta.gov.sg/content/datamall/en/request-for-api.html
2. Register with an email address and verify it.
3. You'll get an AccountKey by email — that's your LTA_API_KEY.

Keep this key server-side only (in the deployment's environment variables). Never put it in the artifact's HTML/JS — anyone viewing the page could read it.

2. Run it locally (optional, to sanity-check)

npm install
cp .env.example .env   # paste your key into .env
npm start

Then:

curl http://localhost:8787/health
# {"ok":true,"hasApiKey":true}

3. Deploy it somewhere public (this repo, already pushed)

It needs a public HTTPS URL — claude.ai connectors are remote MCP servers. Any Node host works; Render's free tier is the least fiddly:

1. In Render: New → Web Service → connect this repo (shiftedtech/parkingpilot).
   • Build command: npm install
   • Start command: npm start
   • Add an environment variable LTA_API_KEY = your key.
2. Deploy. Render gives you a URL like https://parkingpilot.onrender.com.
3. Your MCP endpoint is that URL + /mcp, e.g. https://parkingpilot.onrender.com/mcp.
4. Check https://parkingpilot.onrender.com/health returns {"ok":true,"hasApiKey":true}.

(Render's free tier sleeps after inactivity — the first request after a while takes a few seconds to wake up. Fly.io or Railway work the same way if you'd rather avoid that.)

4. Add it as a claude.ai connector

1. In claude.ai: Settings → Connectors → Add custom connector.
2. Name it something like ParkPilot LTA and paste the /mcp URL from step 3.
3. Save. claude.ai will call tools/list on it and should show carpark_availability.

5. Tell Claude to wire it into the artifact

Once the connector shows up in claude.ai, tell Claude (in the ParkPilot conversation) that it's added — Claude will declare the mcp capability on the artifact with this connector/tool and switch the app over to live HDB carpark data near whatever destination is searched.

Tool reference

carpark_availability(lat?, lng?, radiusKm?, lotType?, limit?)

param	type	notes
lat, lng	number	search center; omit both to get the raw feed sorted by most-available
radiusKm	number	only used with lat/lng, default 2
lotType	"C" | "M" | "H"	car / motorcycle / heavy vehicle, default all
limit	number	max rows returned, default 20, max 200

Returns JSON: { fetchedAt, count, totalMatched, carparks: [{ carParkId, development, area, agency, lotType, availableLots, lat, lng, distanceKm }] }.