Statcan MCP server

StatCan — Statistics Canada

Canada's national statistical office, via the free, keyless Web Data Service (WDS, www150.statcan.gc.ca/t1/wds/rest). Covers the full cube (table) catalogue — population, CPI/inflation, unemployment, GDP, trade, agriculture, and every other official Canadian series — plus headline indicators pre-resolved to a vector id.

Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.

Two access patterns

population by geography) and statcan_series (any series by numeric vector id, FRED-style).

1. Friendly / vector-based — statcan_indicator (cpi, unemployment, gdp,

(dimensions + members for one cube), statcancubevectors (enumerate the actual vector ids for a cube), statcancubedata (latest N observations at a coordinate), statcanchangedseries (daily change feed), statcancsvurl (full-table CSV download link).

1. Cube-based — statcanlistcubes (catalogue), statcancubemetadata

Table number → vector id, in two calls

Every StatCan question starts with a table (e.g. "32-10-0121-01", the egg production table, or its 8-digit form 32100121). Neither the table number nor statcancubemetadata hands you a vector id directly — metadata only gives dimensions and their member lists. Use statcancubevectors to bridge that:

statcan_cube_vectors({ product_id: "32100121", filters: { GEO: "Canada" } })
// -> rows: [{ vector_id: 61133, coordinate: "1.1.0.0.0.0.0.0.0.0",
//            members: { Geography: "Canada", "Production and disposition": "Average number of layers" },
//            frequency: "Monthly", terminated: false, series_title: "Canada;Average number of layers" }, ...]

statcan_series({ vector_id: "61133" })
// -> the actual observations

filters keys match by case-insensitive substring against the cube's dimension names (so GEO matches Geography); values match by substring against member names. Dimensions left unfiltered include every member of that dimension — for a cube with many dimensions that's a large cross-product, so page with limit/page (default 100/page, max 500) rather than pulling everything at once. totalcombinations and hasmore/next_page in the response tell you whether you're seeing all of it.

statcancubemetadata does not return vector ids — it only lists dimensions and members, which is why statcancubevectors exists.

product_id accepts either form

format — the tool strips the dashes and drops the trailing 2-digit suffix)

• The 8-digit WDS product id: 32100121
• The public table number: 32-10-0121-01 (StatCan's NN-NN-NNNN-NN display

Auth

None. Fully keyless — StatCan WDS has no rate-limit key or auth header.

Common pitfalls

one, get it from statcancubevectors first — don't hand-build a coordinate from statcancubemetadata's member list and guess.

• statcancubedata needs a coordinate you already know. If you don't have

dimension (most often GEO) before enumerating a table you haven't seen — otherwise you'll be paging for a long time.

• A wide table has thousands of combinations. Always filter by at least one

stopped publishing; the cube's cubestartdate/cubeenddate describe the whole table's range, not any one series' actual last observation.

• terminated: true on a statcancubevectors row means that series has

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "statcan": {
      "url": "https://gateway.pipeworx.io/statcan/mcp"
    }
  }
}

What this endpoint actually serves

Shortened. The full README is on GitHub.