Statistics Canada MCP Server :: Web Data Service Interface

MCP SDK SDMX REST WDS REST v0.11.0 Python 3.10+ Stateless HTTP

Announcements:
NEW v0.12.0 — data correctness: SDMX data now read as XML; wildcards return every series (previously a partial set), and VECTOR_ID, DGUID and UOM are correct per row
NEW v0.11.0 — statcan CLI for agents: --json envelopes, stable exit codes with hints, statcan describe; fixes for single-period monthly dates and frequency labels
• v0.10.0 — structured output: SDMX tools publish an outputSchema and return typed structuredContent; failures are flagged isError; responses are 25–40% smaller
• v0.9.0 — query_sdmx on this server: fetch a table slice and run one read-only SQL query server-side; only the answer is returned, with no 500-row cap on the fetch
• v0.8.0 — fetch_sdmx_to_database (local install): store a full SDMX slice in SQLite without passing rows through the model; values stored as numbers
• v0.7.16 — statcan download now requires --key; StatCan SDMX no longer accepts the all shortcut
• v0.7.15 — OAuth 2.1/PKCE support: Claude.ai web connector now routes tool calls correctly
• v0.7.14 — fixed 307 redirect loop on POST /mcp; Claude.ai IPs now receive 200 OK
• v0.7.9 — automated CI/CD: version bumps on push, landing announcement workflow
• v0.7.8 — tool-level request logging and /.well-known/mcp.json discovery
• Listed on the MCP server registry

Abstract. This server exposes Statistics Canada’s Web Data Service (WDS) and SDMX REST API as a suite of Model Context Protocol (MCP) tools. Language models connect via the /mcp endpoint and gain structured access to over 7,000 Canadian statistical tables — labour force, GDP, inflation, demographics, health, environment, and more. The server operates in a fully stateless mode: no session state is retained between requests, enabling horizontal scaling and simplified deployment on platforms such as Render.

Connection Instructions

Add this server to your MCP client using the Streamable HTTP transport. No API key is required — Statistics Canada data is publicly available.

MCP Endpoint

https://mcp-statcan.onrender.com/mcp

Health Check

https://mcp-statcan.onrender.com/health

Claude Desktop (`claude_desktop_config.json`)

{
  "mcpServers": {
    "statcan": {
      "type": "http",
      "url": "https://mcp-statcan.onrender.com/mcp"
    }
  }
}

Claude Code (`.mcp.json`)

{
  "mcpServers": {
    "statcan": {
      "type": "http",
      "url": "https://mcp-statcan.onrender.com/mcp"
    }
  }
}

Claude Code (command line)

claude mcp add --transport http statcan https://mcp-statcan.onrender.com/mcp

Local install (stdio, includes SQLite tools and the `statcan` CLI)

pip install statcan-mcp-server

CLI for agents with a shell (JSON output, stable exit codes)

statcan describe

statcan --json download 18-10-0004-01 --key "2+3.2" --last 12 --output cpi.csv

Note: This server is deployed on Render’s free tier. The first request after a period of inactivity may take 30–60 seconds while the instance spins up. Subsequent requests are fast.

Command-Line Interface

The statcan CLI ships with the local install and works without an LLM. People can browse tables in the terminal; agents with a shell can use --json for structured output and stable exit codes.

Exploring a table: search, metadata, download

Terminal recording: statcan search, statcan metadata and statcan download for the Consumer Price Index table

statcan search → statcan metadata → statcan download (table output)

Agent mode: JSON envelopes and exit codes

Terminal recording: statcan --json download writes a CSV and prints a summary envelope; a missing product returns a JSON error with exit code 3; statcan describe lists exit codes

--json download writes the file and prints a summary; failures return a typed exit code and a hint

Registered Tools (18)

All tools below are exposed over the /mcp endpoint. The hosted HTTP server runs in stateless mode — composite and SQLite database tools are intentionally omitted here. To use those, run the server locally over stdio (pip install statcan-mcp-server). For calculations on this server, use query_sdmx: the SQL runs here and only the answer is sent back to the model.

Tool Name	Category	Description
search_cubes_by_title	Discovery	Full-text keyword search across all StatCan tables, capped at 25 results
get_all_cubes_list	Discovery	Paginated inventory of all tables (100/page)
get_all_cubes_list_lite	Discovery	Lightweight paginated list, 1-hour TTL cache
get_cube_metadata	Metadata	Full table structure: dimensions, members, frequencies; summary mode caps members at 10
get_code_sets	Metadata	WDS attribute decoder tables (scalar type, UOM, frequency, status codes)
get_series_info	Series	Resolve dimension coordinates to vectorId, frequency, UOM
get_series_info_from_vector	Series	Look up series metadata by vectorId
get_bulk_vector_data_by_range	Vector	Fetch multiple vectors filtered by release-date range
get_changed_series_data_from_vector	Change Detection	Detect data changes for a vector since a given date
get_changed_series_data_from_cube_pid_coord	Change Detection	Detect changes for a specific series by coordinate
get_changed_cube_list	Change Detection	List all tables with data changes since a given date
get_changed_series_list	Change Detection	List all series with changes since a given date
get_sdmx_structure	SDMX	Data Structure Definition as JSON: dimension positions and codelists with hierarchy
get_sdmx_data	SDMX	Server-side filtered observations by productId + SDMX key string
get_sdmx_rows	SDMX	Inline tabular rows (up to 500) returned directly in the tool response
query_sdmx	SDMX + SQL	Fetch observations and run one read-only SQL query server-side in a temporary in-memory table; returns only the result (rank, compare, growth)
get_sdmx_vector_data	SDMX	Single vectorId observations via SDMX
get_sdmx_key_for_dimension	SDMX	Returns the full OR-key for a large dimension (avoids wildcard sampling errors)

Technical Specifications

Transport	Streamable HTTP (stateless) — MCP 2025-03-26 spec
Upstream APIs	Statistics Canada WDS REST + SDMX 2.1 REST (data read as structure-specific XML)
Authentication	OAuth 2.1/PKCE available (auto-approved, no credentials); open access also supported
SSL Verification	On by default; `STATCAN_VERIFY_SSL=false` disables it for broken trust stores
Structured Output	`outputSchema` + `structuredContent` on all SDMX tools; read-only tools carry `readOnlyHint`
Max SDMX Rows	500 per `get_sdmx_data` call
Server-side SQL	`query_sdmx`: SQLite in-memory, one SELECT per call, 2 s limit, up to 50,000 fetched rows, 500 result rows; nothing stored
Cache	1-hour TTL on `get_all_cubes_list_lite`
Python	3.10+ — asyncio, httpx, Starlette, Uvicorn
Install	`pip install statcan-mcp-server`
Source	github.com/Aryan-Jhaveri/mcp-statcan

Known Constraints & Notes

lastNObservations + startPeriod/endPeriod — cannot be combined; StatCan returns HTTP 406.
Wildcards include totals — . returns every member of a dimension, including totals and parent aggregates. For sums or rankings use leaf codes from get_sdmx_key_for_dimension.
vectorIds from metadata — get_cube_metadata returns member names but not vectorIds. Use get_series_info to resolve coordinates to vectorIds.
Data format — SDMX data is read as structure-specific XML and labelled from the table’s structure. StatCan’s SDMX-JSON mis-indexes series and attributes, which previously caused missing wildcard rows, wrong OR-key labels and misplaced VECTOR_ID/UOM values (fixed in v0.12.0).

Data Attribution

All data returned by this server originates from Statistics Canada. Users are responsible for complying with Statistics Canada’s Open Licence when using or redistributing data obtained through this server. This MCP server is an independent project and is not affiliated with or endorsed by Statistics Canada or the Government of Canada.

Statistics Canada MCP Server