StatsMapped Public Data

What the StatsMapped Public Data MCP server does

The StatsMapped Public Data MCP server gives AI agents access to public statistics maintained by StatsMapped. Coverage includes Ireland by county and the UK by local authority, with data spanning housing, crime, health, the economy, and social welfare. The underlying figures come from official publishers, including Ireland’s CSO, PSRA, Central Bank of Ireland, DHLGH, NTPF, the Office of Government Procurement, and the EU Publications Office, along with UK sources such as ONS, HM Land Registry, Nomis, DfE, and DfT.

Each result can include source information and caveats. The package is a thin client: it sends requests to StatsMapped’s unauthenticated public HTTPS API and does not store or modify data. No API key is needed.

How it works

The StatsMapped Public Data MCP server normally starts on the user’s machine and communicates with an MCP client over stdio. Tool calls then make HTTPS requests to the public StatsMapped API. This arrangement avoids routing every user through one hosted server and is the recommended operating mode in the project documentation.

A Streamable HTTP mode is also available by setting the transport configuration for a deployment. A hosted endpoint is available at https://mcp.statsmapped.com/mcp, but remote users share the endpoint’s source IP when it proxies API requests. The public API is rate-limited to 600 requests per hour per source IP, so deployment architecture affects how that limit is shared.

Every tool accepts country, with ireland as the default and united-kingdom as the other supported value. Geography and dataset availability differ between the two countries. Irish examples use county identifiers such as county:kerry; UK examples use local authority identifiers such as uk:lad:e09000033.

Setup and configuration

Install the package from PyPI with:

pip install statsmapped-mcp

Add the executable to an MCP client configuration. For example:

{
  "mcpServers": {
    "statsmapped": {
      "command": "statsmapped-mcp"
    }
  }
}

Claude Desktop uses its claude_desktop_config.json file, while Cursor uses an equivalent mcp.json configuration. The default setup needs no API key or required environment variable. The API base URL can be overridden with STATSMAPPED_MCP_BASE_URL; the development tests otherwise use the live StatsMapped API.

Tools and capabilities

The StatsMapped Public Data MCP server provides four tools:

• query_data lists all tracked statistics for a country, lists datasets available for an area with the latest value and year-on-year change, or returns full detail for one area and dataset. The detailed mode can include recent history through history_months.
• list_areas returns geographies for a boundary level. Ireland defaults to county; the UK’s primary level is lad, meaning local authority district.
• compare ranks areas by a statistic, returns details for a registered comparison pair, checks whether two statistics have a registered comparison, or lists all registered pairs. It does not calculate arbitrary new correlations.
• explain_metric returns a statistic’s definition, methodology, and standing caveats, rather than its current value.

Use query_data without an area or dataset to discover available statistics before making a more specific request. Comparison checks return yes, no, or unknown as a string, so clients should compare the value explicitly rather than rely on truthiness.

Limitations and notes

Coverage is limited to the datasets and geography levels tracked by StatsMapped. Ireland and the UK do not expose identical datasets or boundary structures. Comparison functionality is registry-backed and hand-vetted; most arbitrary statistic pairs return unknown, and no fresh correlation is calculated.

The API is read-only for this package. Statistics retain their publishers’ licensing terms, so users should review the relevant source and licence information before reusing figures outside an agent query. Full caveat text is returned for a specific area-and-dataset request; area-level dataset listings provide only a caveat label and severity. The package is MIT-licensed, but that licence does not replace the licences attached to the underlying data.