ausdata.io

The Model Context Protocol (MCP) is Anthropic open standard for connecting LLM-powered agents to live tools. The ausdata MCP server exposes 9 Australian government data sources as MCP tools. Install it once in your host (Claude Desktop, Cursor, Windsurf) and every agent can call live AU data inline.

Install in 30 seconds

The MCP server is a single npm package:

npx ausdata-mcp

You do not need to install it persistently --- npx runs it on demand. For configuration, you reference it from your host MCP config file.

Claude Desktop setup

Open Claude Desktop, then Settings → Developer → Edit Config. Add:

{
  "mcpServers": {
    "ausdata": {
      "command": "npx",
      "args": ["ausdata-mcp"],
      "env": {
        "AUSDATA_API_KEY": "ak_YOUR_KEY"
      }
    }
  }
}

Restart Claude Desktop. The ausdata tools appear in the available tools list.

Cursor setup

In Cursor: Settings → MCP → Add new server:

{
  "name": "ausdata",
  "command": "npx",
  "args": ["ausdata-mcp"],
  "env": {
    "AUSDATA_API_KEY": "ak_YOUR_KEY"
  }
}

Windsurf setup

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "ausdata": {
      "command": "npx",
      "args": ["ausdata-mcp"],
      "env": {
        "AUSDATA_API_KEY": "ak_YOUR_KEY"
      }
    }
  }
}

Get an API key first

curl -X POST https://api.ausdata.io/v1/register \
  -H "Content-Type: application/json" \
  -d '{"email":"you@example.com"}'

Free tier: 500 calls per month, no card.

Tools the server exposes

The MCP server exposes 46 tools across 9 sources. The five core tools per source follow the standard pattern:

  • {source}_search_datasets(query) --- fuzzy search
  • {source}_describe(dataset_id) --- schema + filters
  • {source}_get_data(dataset_id, filters, start_period, end_period) --- fetch records
  • {source}_latest(dataset_id, filters) --- current snapshot
  • {source}_list_curated() --- enumerate available datasets

Plus the 23 composer endpoints exposed as additional tools (real_rate_regime, inflation_decomposition, nem_energy_regime, etc.).

Example agent interactions

"What is the current cash rate?"

The agent calls:

tool: rba_get_data
args: {"dataset_id": "F1", "filters": {"series": "cash_rate_target"}, "limit": 1}

Returns the current cash rate value with the announcement date.

"Compare CPI by capital city"

The agent calls:

tool: cpi_by_city
args: {}

Returns annual change percentage for all 8 capital cities × 3 expenditure categories.

"Is the real cash rate positive right now?"

The agent calls:

tool: real_rate_regime
args: {}

Returns nominal cash rate, trimmed-mean CPI, real cash rate, and regime classification (positive / neutral / negative). The agent does not need to learn the math.

Why this matters

Most LLMs are trained 6-18 months before deployment. Australian macro data (cash rate, CPI, employment) changes regularly. An agent answering AU questions from training data will be wrong sooner or later. The MCP server eliminates this failure mode by giving the agent direct access to live data.

This matters operationally for:

  • Financial advisor copilots
  • Loan / mortgage comparison agents
  • AI research assistants
  • Property valuation tools
  • Energy market analysts

For agents in these spaces, "the cash rate is currently 3.85%" when the actual rate is 4.10% is a customer-facing reliability problem. Grounded data via MCP eliminates it.

The 9 sources covered

The MCP server includes sister modules for:

  • abs --- Australian Bureau of Statistics (CPI, labour force, wages, GDP, etc.)
  • rba --- Reserve Bank of Australia (cash rate, lending rates, credit)
  • ato --- Australian Taxation Office (postcode income, occupation, ACNC)
  • apra --- Australian Prudential Regulation Authority (banking, super, insurance)
  • aihw --- Australian Institute of Health and Welfare (mortality, hospitals)
  • asic --- Australian Securities and Investments Commission (companies, registers)
  • aemo --- Australian Energy Market Operator (NEM dispatch, generation)
  • wgea --- Workplace Gender Equality Agency (gender pay gap by industry)
  • au_weather --- Bureau of Meteorology via Open-Meteo (weather by city)

Rate limits and cost

The MCP server inherits the free / paid tier of your underlying API key:

  • Free (500 calls/month, 10/min): plenty for personal use and evaluation
  • Analyst ($29/mo, 10k calls, 60/min): right for development
  • Embed ($99/mo, 100k calls, 300/min): right for production agents

For most production AI applications doing ~10-50 grounded queries per session × 100-500 sessions per day, the Embed tier is the right shape.

Source

The MCP server is open source on GitHub at Bigred97/ausdata-mcp. Reports issues, suggest features, contribute datasets you want covered.

Common gotchas

  1. Forgot to set AUSDATA_API_KEY env var. Every tool call returns 401. Fix: add the env var to the MCP config.
  2. Restart required after config change. Claude Desktop / Cursor / Windsurf re-read the MCP config only on start.
  3. First call cold latency. First call to an uncommon dataset may take 5-15 seconds while our cache warms. Subsequent calls are sub-second.
  4. Rate limits. The free tier caps at 10 requests per second. Burst above that, the agent gets a 429 with a retry-after header.

Related

  • /docs/mcp --- full reference
  • /how-to/connect-claude-to-australian-data --- install walkthrough
  • /how-to/prevent-llm-hallucinating-cash-rate --- grounding patterns
  • /for/ai-agents --- agent-builder landing page

Get a free API key → /register

Get a key and call your first endpoint

500 free calls per month. No credit card.

Get a free API key