MonstarX Singapore MCP

A Model Context Protocol (MCP) server that gives AI agents and applications live, structured access to Singapore government open data — weather, transport, carparks, property prices, company registry, geocoding, and more.

It wraps official public APIs (data.gov.sg, LTA DataMall, OneMap/SLA, NEA, ACRA, HDB, MOE) behind 35 clean, well-described tools with a consistent, provenance-stamped response format. No API keys or authentication are required for clients — the server handles upstream credentials for you.

🎮 Live playground: https://sg-mcp-playground-production.up.railway.app — pick a tool, tweak the inputs, hit Run, and see real Singapore data render as maps, weather cards, tables and JSON, right in your browser.

🔌 MCP endpoint (staging): https://sg-mcp-staging.monstarxapp.com/mcp Server version 0.1.0 · Protocol: MCP · Transport: Streamable HTTP

This repository contains the playground / documentation site (a self-contained static page) and a tiny server to host it. See Run locally & deploy.

Table of contents

• Run locally & deploy

• Repository layout

• Quick start

• Connecting from MCP clients

  • Claude Code

  • Claude Desktop

  • Cursor / Windsurf / other MCP clients

• Endpoints

• Transport & protocol details

• Response format

• Tool reference

  • Open data catalog

  • Weather & environment

  • Carparks

  • Public transport

  • Geocoding & addresses

  • Property

  • Companies (ACRA)

  • Education (Graduate Employment Survey)

  • Public health

• Example prompts for AI agents

• Code examples

• Error handling

• Rate limits, caching & data freshness

• Data sources & licensing

Related MCP server: OneMap MCP Server

Run locally & deploy

The site is a single self-contained HTML file (public/index.html) served by a tiny dependency-free Node server.

npm start           # serves public/ on http://localhost:8080  (respects $PORT)

Because the MCP server sends Access-Control-Allow-Origin: *, the in-browser playground calls it directly — no backend needed.

Deploy to Railway:

railway up --ci     # uses railway.json (Nixpacks build, /health healthcheck)

Regenerate the page after editing content (the HTML is generated from build/):

python3 build/build.py     # rewrites public/index.html + singapore-mcp-playground.html

Repository layout

├── public/index.html            # the playground (what gets served)
├── singapore-mcp-playground.html # standalone copy (identical; open locally)
├── server.js                    # dependency-free static server, respects $PORT, /health
├── package.json                 # start script, Node ≥18
├── railway.json                 # Railway build + healthcheck config
├── build/
│   ├── build.py                 # generator for the playground HTML
│   └── data.min.json            # tool metadata + captured example responses
├── examples/                    # standalone MCP clients (Python, TS, curl, SDK)
└── README.md

Quick start

The server speaks JSON-RPC 2.0 over HTTP. You can talk to it with nothing but curl.

1. Discover the server:

curl https://sg-mcp-staging.monstarxapp.com/

2. List available tools:

curl -X POST https://sg-mcp-staging.monstarxapp.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-protocol-version: 2025-06-18" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

3. Call a tool — how many taxis are available in Singapore right now?

curl -X POST https://sg-mcp-staging.monstarxapp.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-protocol-version: 2025-06-18" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"sg_taxi_availability","arguments":{}}}'

{
  "result": {
    "content": [{ "type": "text", "text": "{ ...json... }" }],
    "structuredContent": {
      "source": "data.gov.sg",
      "retrieved_at": "2026-07-22T02:06:41.665Z",
      "license": "Singapore Open Data Licence",
      "api": "real-time taxi availability",
      "agency": "LTA",
      "data": { "available_taxis": 3190, "timestamp": "2026-07-22T10:06:23+08:00" }
    }
  },
  "jsonrpc": "2.0",
  "id": 2
}

That's it — no auth, no session setup. See Code examples for Python and TypeScript clients.

Connecting from MCP clients

This is a remote (HTTP) MCP server. Most MCP clients can connect either natively (if they support remote HTTP servers) or through the mcp-remote bridge.

Claude Code

claude mcp add --transport http sg-mcp https://sg-mcp-staging.monstarxapp.com/mcp

Then in a session: "Using the sg-mcp tools, what's the 2-hour weather forecast for Tampines?"

Claude Desktop

Edit your claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "singapore": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://sg-mcp-staging.monstarxapp.com/mcp"]
    }
  }
}

Restart Claude Desktop; the 35 sg_* tools appear in the tools menu.

Cursor / Windsurf / other MCP clients

Any client that supports remote HTTP MCP servers can use the URL directly:

{
  "mcpServers": {
    "singapore": {
      "type": "http",
      "url": "https://sg-mcp-staging.monstarxapp.com/mcp"
    }
  }
}

For clients that only support stdio, use the mcp-remote bridge shown in the Claude Desktop example.

Endpoints

CORS is open (Access-Control-Allow-Origin: *), so browser-based clients work too.

Transport & protocol details

• Transport: Streamable HTTP (MCP spec). POST JSON-RPC to /mcp.

• Protocol version: 2025-06-18 (send it in the mcp-protocol-version header).

• Stateless: the server does not issue an mcp-session-id. You do not need to call initialize before tools/call when using raw HTTP — each request is independent. (Full MCP clients will still perform the initialize handshake, which the server supports.)

• Accept header: include application/json, text/event-stream.

• Auth: none required.

Minimal request skeleton:

POST /mcp HTTP/1.1
Host: sg-mcp-staging.monstarxapp.com
Content-Type: application/json
Accept: application/json, text/event-stream
mcp-protocol-version: 2025-06-18

{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"<tool>","arguments":{...}}}

Response format

Every successful tool call returns an MCP result with two mirrored representations of the same payload:

• content[0].text — the payload as a JSON string (for text-only clients / LLMs).

• structuredContent — the same payload as a native JSON object (prefer this when parsing programmatically).

Every payload uses a consistent provenance envelope:

{
  "source":      "data.gov.sg",                    // upstream platform
  "retrieved_at":"2026-07-22T02:06:41.665Z",       // server fetch time (UTC)
  "license":     "Singapore Open Data Licence",    // when applicable
  "agency":      "LTA",                            // originating gov agency
  "api":         "real-time taxi availability",    // upstream API used
  "data":        { ... }                           // the actual result
}

Some tools add top-level context fields (e.g. query, total, shown, found, dataset_id) alongside data/results/companies. Timestamps in the envelope are UTC; live data timestamps from LTA/NEA are typically SGT (+08:00).

Tool reference

All 35 tools are prefixed sg_. Required parameters are marked *.

Open data catalog

Generic access to any tabular dataset on data.gov.sg — the escape hatch when no dedicated tool exists.

Weather & environment

All are real-time NEA feeds and take no parameters.

Carparks

HDB carpark information and real-time lot availability.

Public transport

LTA DataMall real-time and reference transport data.

Geocoding & addresses

OneMap (Singapore Land Authority) address search and geocoding.

Property

Companies (ACRA)

Singapore's registry of business entities.

Education (Graduate Employment Survey)

MOE Graduate Employment Survey — salaries and employment outcomes by degree.

Public health

Example prompts for AI agents

Natural-language prompts an agent can satisfy using these tools:

• "Is it going to rain in Jurong in the next 2 hours?" → sg_weather_2h

• "What's the UV index right now — do I need sunscreen?" → sg_uv_index

• "Find HDB carparks near Tampines with available lots right now." → sg_carpark_search (include_availability:true)

• "When's the next bus 15 at stop 83139?" → sg_bus_arrival

• "Are there any train delays this morning?" → sg_train_service_alerts

• "Any traffic incidents on the expressways right now?" → sg_traffic_incidents

• "What did 4-room flats in Tampines sell for recently?" → sg_hdb_resale_prices

• "Which NUS degrees had the highest median starting salary in 2022?" → sg_ges_top_degrees

• "Is 'DBS Bank Ltd' a registered company? What's its UEN?" → sg_company_verify / sg_company_search

• "Get the coordinates of ION Orchard (postal 238801)." → sg_geocode

• "What's near latitude 1.2834, longitude 103.8607?" → sg_reverse_geocode

• "Where are the active dengue clusters with 10+ cases?" → sg_dengue_clusters

System-prompt hint for agents: "You have access to the Singapore MCP server (sg_* tools) for live Singapore data — weather, transport, carparks, HDB prices, companies, geocoding. Prefer the specific tool over the generic sg_dataset_query. Always cite the source/agency and retrieved_at from responses when reporting live figures."

Code examples

Runnable clients live in examples/:

• examples/curl.sh — pure shell / curl.

• examples/client.py — Python (stdlib only, no dependencies).

• examples/client.ts — TypeScript / Node (fetch).

• examples/mcp_sdk_client.py — using the official mcp Python SDK.

Python (no dependencies)

import json, urllib.request

BASE = "https://sg-mcp-staging.monstarxapp.com/mcp"

def call_tool(name, arguments=None):
    body = json.dumps({
        "jsonrpc": "2.0", "id": 1, "method": "tools/call",
        "params": {"name": name, "arguments": arguments or {}},
    }).encode()
    req = urllib.request.Request(BASE, data=body, headers={
        "Content-Type": "application/json",
        "Accept": "application/json, text/event-stream",
        "mcp-protocol-version": "2025-06-18",
    })
    with urllib.request.urlopen(req) as r:
        result = json.load(r)["result"]
    if result.get("isError"):
        raise RuntimeError(result["content"][0]["text"])
    return result["structuredContent"]

taxis = call_tool("sg_taxi_availability")
print("Available taxis:", taxis["data"]["available_taxis"])

carparks = call_tool("sg_carpark_search",
                     {"query": "TAMPINES", "include_availability": True, "limit": 3})
print(json.dumps(carparks, indent=2))

TypeScript / Node

const BASE = "https://sg-mcp-staging.monstarxapp.com/mcp";

async function callTool(name: string, args: Record<string, unknown> = {}) {
  const res = await fetch(BASE, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Accept: "application/json, text/event-stream",
      "mcp-protocol-version": "2025-06-18",
    },
    body: JSON.stringify({
      jsonrpc: "2.0", id: 1, method: "tools/call",
      params: { name, arguments: args },
    }),
  });
  const { result } = await res.json();
  if (result.isError) throw new Error(result.content[0].text);
  return result.structuredContent;
}

const w = await callTool("sg_weather_2h");
console.log(w.data.items?.[0] ?? w.data);

Error handling

Errors are returned as a normal JSON-RPC result with isError: true (not as a JSON-RPC top-level error). The message is in content[0].text.

Invalid / missing arguments → MCP error -32602:

{
  "result": {
    "content": [{ "type": "text",
      "text": "MCP error -32602: Input validation error: Invalid arguments for tool sg_company_by_uen: [{\"expected\":\"string\",\"path\":[\"uen\"],\"message\":\"expected string, received undefined\"}]" }],
    "isError": true
  }
}

Unknown tool → MCP error -32602: Tool <name> not found.

No matching data is not an error — you get a normal envelope with an empty data/results array (e.g. querying an unknown bus stop returns "Services": []). Always check for empty results in addition to isError.

Handling pattern:

if result.get("isError"):
    # inspect result["content"][0]["text"]
    ...
else:
    payload = result["structuredContent"]
    if not payload.get("data") and not payload.get("results"):
        # valid call, no matches
        ...

Rate limits, caching & data freshness

• Freshness: real-time feeds (weather, taxi, bus arrivals, carpark availability, traffic) reflect the upstream agency's latest publish. Check retrieved_at (server fetch, UTC) and the upstream timestamp (usually SGT) in the payload.

• Reference data (bus stops/services/routes, carpark info, datasets) changes rarely; safe to cache client-side.

• Rate limits: the server proxies rate-limited government APIs (notably LTA DataMall). Be a good citizen — cache reference data, avoid tight polling loops, and back off on errors. This is a staging deployment; do not use it for production load.

Data sources & licensing

Data is sourced from official Singapore government platforms and remains subject to their terms:

Most datasets are provided under the Singapore Open Data Licence (echoed in each response's license field). LTA DataMall and OneMap data are subject to their respective terms of use. You are responsible for complying with the source licences when using or redistributing the data. This MCP server is an independent wrapper and is not endorsed by any government agency.

Server: Monstarx Singapore MCP v0.1.0 · staging · This documentation was generated by testing the live endpoint on 2026-07-22.