MapleStats MCP

One MCP server for Canadian open data.

MapleStats MCP gives AI agents (Claude, Cursor, and any MCP-compatible client) structured, typed access to Canadian open data through a single server — covering Statistics Canada (tables, Census and public use microdata), the Bank of Canada, CMHC, federal agencies, and federal, provincial, territorial and municipal open-data portals, in English and French.

MapleStats MCP donne aux agents IA (Claude, Cursor et tout client compatible MCP) un accès structuré et typé aux données ouvertes canadiennes par l'entremise d'un seul serveur — couvrant Statistique Canada, la Banque du Canada, ainsi que des portails CKAN fédéraux, provinciaux, territoriaux et municipaux vérifiés.

See PROJECT_GUIDE.md for the project vision and ROADMAP.md for source coverage status.

The easiest way: ask your agent

1. Copy this prompt.
2. Paste it into Claude Code, Codex, Cursor or any agent that can run commands on your computer.
3. Restart the agent when it says so, then ask for data.

Install the MapleStats MCP server and connect it to this agent. Follow the setup steps in https://github.com/dsanchezp18/maplestats-mcp

Prefer to do it by hand? See Install locally or the Connect page.

Instructions for the agent doing the setup

1. Make sure uv is installed (uv --version). If it is missing, install it with the official installer from that page.

2. Use this launch command. It installs straight from GitHub and works before and after the PyPI release: uvx --from git+https://github.com/dsanchezp18/maplestats-mcp maplestats-mcp.

3. Register it with the client you are running in, under the name maplestats:

   • Claude Code: claude mcp add --scope user maplestats -- uvx --from git+https://github.com/dsanchezp18/maplestats-mcp maplestats-mcp
   • Codex CLI: codex mcp add maplestats -- uvx --from git+https://github.com/dsanchezp18/maplestats-mcp maplestats-mcp
   • Cursor, Claude Desktop and other mcpServers clients: add the entry below to the client's config file (Cursor: ~/.cursor/mcp.json; Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows), keeping any servers already there.
   • VS Code: the same entry under a servers key (not mcpServers) in .vscode/mcp.json.

   {
     "mcpServers": {
       "maplestats": {
         "command": "uvx",
         "args": ["--from", "git+https://github.com/dsanchezp18/maplestats-mcp", "maplestats-mcp"]
       }
     }
   }
   

4. The server speaks MCP over stdio and needs no API key, account or open port. Tell the user to restart or reload the client, then check that the search_tools tool is listed.

More clients and hosting options: Connect.

Bilingual by design / Conçu pour être bilingue

Tools accept lang: "en"|"fr", several sources are French-first or French-only (Quebec's and Montreal's CKAN catalogues, ckan_* with portal="qc"/"montreal"), and tool discovery works in either language: every tool's docstring carries both a Keywords: line and a Mots-clés : line, so calling search_tools with a French-language query (e.g. "recherche de jeux de données sur le climat") finds the same tools an equivalent English query would. Read docs://catalogue for a bilingual (EN/FR) one-line description of every module. Not every underlying government portal is itself bilingual — see each module's own docstring (or docs://catalogue) for where lang genuinely changes the response versus where it is a documented no-op on a monolingual source.

Chaque outil accepte lang : "en"|"fr", plusieurs sources sont francophones ou exclusivement en français (les catalogues CKAN du Québec et de Montréal, ckan_* avec portal="qc"/"montreal"), et la découverte d'outils fonctionne dans les deux langues : chaque outil porte à la fois une ligne Keywords: et une ligne Mots-clés :, de sorte qu'un appel à search_tools avec une requête en français trouve les mêmes outils qu'une requête équivalente en anglais. Consultez docs://catalogue pour une description bilingue (EN/FR) de chaque module. Tous les portails gouvernementaux sous-jacents ne sont pas eux-mêmes bilingues — consultez la documentation de chaque module (ou docs://catalogue) pour savoir où lang change réellement la réponse et où il s'agit d'un no-op documenté sur une source unilingue.

Status

About 200 tools across these sources (run docs://catalogue for a bilingual one-line description of each module):

Many other federal administrative series (CRA statistics and charities, OSFI returns, ISED insolvency data) are ordinary open.canada.ca datasets, reachable through ckan_search_datasets(portal="federal", fq="organization:<org>").

Most tools accept lang: "en"|"fr" (a documented no-op on single-language sources), and every tool returns a typed response with a provenance block (source, URL, query time, freshness, limits). See AGENTS.md for the full architecture and response contract.

Install locally (no Docker required)

MapleStats MCP is not on PyPI yet, so install it from GitHub for now. With uv, MCP clients can run it without a separate install step:

uvx --from git+https://github.com/dsanchezp18/maplestats-mcp maplestats-mcp

Or install the command once:

uv tool install git+https://github.com/dsanchezp18/maplestats-mcp.git

pip install git+https://github.com/dsanchezp18/maplestats-mcp.git

The first PyPI release will publish it as maplestats-mcp. From then on the shorter forms work too: uvx maplestats-mcp, uv tool install maplestats-mcp and pip install maplestats-mcp.

Once installed, the command is on your PATH:

maplestats-mcp

It speaks MCP over stdio by default, which is the format local MCP clients expect. The process is started by the client; do not open a port and do not run Docker.

To update an existing installation:

uv tool upgrade maplestats-mcp

If you are working from a clone instead:

uv sync
uv run maplestats-mcp

MCP client configuration

For clients that accept a standard mcpServers JSON configuration, add:

{
  "mcpServers": {
    "maplestats": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/dsanchezp18/maplestats-mcp", "maplestats-mcp"]
    }
  }
}

For Claude Code:

claude mcp add --scope user maplestats -- uvx --from git+https://github.com/dsanchezp18/maplestats-mcp maplestats-mcp

After the PyPI release, "args": ["maplestats-mcp"] and uvx maplestats-mcp are enough. If you installed the command with uv tool install or pip, use "command": "maplestats-mcp" with no args instead of uvx.

On Windows, make sure the directory where uv installs tools is on PATH, then restart the MCP client after installation.

Development quick start

uv sync

Run directly from a checkout (stdio, for local MCP clients):

uv run maplestats-mcp

Run as a hosted HTTP server:

MAPLE_TRANSPORT=http MAPLE_HOST=0.0.0.0 MAPLE_PORT=8000 uv run maplestats-mcp

See Hosting below for the full environment-variable surface (auth, rate limiting, TLS).

Using it

Tools are discovered through a search layer rather than listed flat — call search_tools with a plain-language query, then call_tool with the name it returns:

{"name": "search_tools", "arguments": {"query": "consumer price index"}}
{"name": "call_tool", "arguments": {"name": "wds_search_cubes", "arguments": {"query": "consumer price index"}}}

Two example workflows, also available as guided MCP prompts (find_and_fetch_series, look_up_classification, build_sdmx_or_key):

• Find and fetch a data series: wds_search_cubes → wds_get_cube_metadata → wds_get_series_info_from_cube_pid_coord → wds_get_data_from_vectors.
• Look up a classification: rdaas_search_classifications → rdaas_get_classification → rdaas_get_classification_categories_detailed.

Development

uv sync                          # install
uv run ruff check src tests      # lint
uv run ruff format src tests     # format
uv run pyright                   # type check
uv run pytest                    # unit tests (mocked, no network)

Live verification (hits the real StatCan APIs — needs outbound HTTPS, not just mocks):

.\scripts\verify.ps1

runs the full gate above plus every scripts/smoke_test*.py live smoke test. If Docker is installed, it also runs a build, compose up, and health check.

See AGENTS.md for the full contributor guide, including how to add a new source module.

Website: site/ holds the project website, generated from the tool registry by scripts/build_site.py:

uv run python scripts/build_site.py              # writes build/site/
uv run python -m http.server --directory build/site 8080

Hosting

MAPLE_TRANSPORT=http MAPLE_REQUIRE_AUTH=0 docker compose up --build

Docker is an optional deployment method. For a personal computer, use the local installation above. If the HTTP server is exposed beyond the local machine, set MAPLE_AUTH_TOKEN and keep MAPLE_REQUIRE_AUTH=1.

GET /health reports uptime and version; it bypasses auth/rate limiting so it's always reachable.

License

MIT

Acknowledgments

The module architecture (per-source folders, auto-registered tools, bilingual response envelope) and the hosting layer (Bearer auth, sliding-window rate limiting, health checks) were informed by prior open-source work building MCP servers for government and public data — most directly ReyemTech's mcp-canada for the module pattern and DweskZ's EcuDataMCP for the hosting middleware, alongside the StatCan-specific benchmarks reviewed while researching this project. Thank you to their authors for building in the open.

Much of the Canadian data work here stands on Jens von Bergmann's (mountainMath) R packages and their co-authors:

• cmhc: the CMHC module was audited against its reverse-engineering of CMHC's Housing Market Information Portal.
• cansim: the R code that reproduce_code generates for Statistics Canada tables and vectors uses it.
• cancensus: a model for census data access in R.
• canivt: reads Beyond 20/20 IVT files; census tables and Borealis deposits that exist only in that format are routed to it, and its sampling notes pointed to Borealis as the main IVT holder outside StatCan.

Alternatives

MapleStats is not the only way to reach this data. The website's About page lists the alternatives: other MCP servers for Canadian data, R and Python packages, and the official APIs MapleStats calls.