Questions people ask

No. The code stores no queries, has no accounts and sends nothing to any analytics service.

The one thing it keeps is a count of how many times each tool has run, by the tool's name, the day and whether the call worked; the counts sit in memory and are public at /stats on the hosted server. The server does see the tool calls your AI agent makes, such as a search phrase or a table number, because it has to run them, but never your conversation, and it writes none of them down: its console log shows start-up messages and, over HTTP, the method, path and status of each request, never what is in them. This is about the code; if you use the hosted server, read the answer on it below.

Two things are kept to save time: an in-memory copy of recent responses, gone when the server stops, and the microdata files, patent tables and StatCan open-database files that some tools download, kept in cache folders you choose. Neither holds your questions. On your own machine, reproduce_workbook also saves the Excel workbooks you ask for, in your Downloads folder unless you choose another.

Your AI assistant handles your conversation under its own privacy policy. When a tool runs, the publisher's service, such as Statistics Canada, the Bank of Canada or a city's portal, sees that request and the address it comes from, as with any web page. If you use the hosted endpoint, the server and its host, Render, are on that path too. MapleStats itself stores nothing.

Yes. A public endpoint runs at https://maplestats-mcp.onrender.com/mcp, with no account or API key. It is a free instance: a ping every 10 minutes keeps it awake, so cold starts are rare, but the first request after an idle spell can take up to a minute. Each client is limited to 60 requests a minute, and the public use microdata tabulation tool (statcan_pumf_tabulate) is switched off there because it downloads whole files. For those, or for anything you would rather keep on your own machine, run MapleStats locally.

On the hosted endpoint your tool calls do reach the server, since it has to run them. The server code does not store or log them: its access log records the method, path and status of each request, not what is in it, the usage counts hold tool names only (never arguments or client addresses), and the rate limiter keeps a client's address in memory for a minute to count requests. Render, the host, keeps its own request logs under its own terms, so read Render's terms and privacy policy before you rely on it. If you want full privacy, run MapleStats locally: then nothing leaves your computer except the requests to the publishers.

It comes live from the publisher, kept for a short time in memory. Each result says where it came from, when it was queried and any limits, in its provenance block.

Yes. Before a source is added, its terms of use are read, and a source whose terms ban automated use is left out.

Some useful portals are therefore not covered. Examples are Canada's Business Registries, SEDAR+, Payments Canada and CAPP's Statistics Handbook, whose terms forbid automated tools, scraping or bulk use. Petrinex's public oil and gas data is not built either, because the host that serves its downloads does not permit automated access. Some sources were removed after a later review of their terms: CanadaBuys procurement data, OpenParliament.ca and the Toronto Transit Commission's own schedule download; the Saskatchewan GeoHub portal was also removed. Justice Laws (federal statutes and regulations) is out of scope, because legislation text is not data this server carries. The Saskatchewan Bureau of Statistics' Provincial Economic Accounts are not built: the tables could be fetched, but they are under Crown copyright for non-commercial reuse only and the site's terms bar automated access, which a public server cannot keep to. The Toronto Transit Commission's schedule is still available from Statistics Canada's national transit database. Sites that block automated requests are also left alone. Each skipped source and the reason are recorded in the roadmap. Where a source asks for credit or a notice, the result's provenance block names the publisher so you can give it.

Yes, for most data calls. Ask for the script: reproduce_code writes R, Python, Stata or Julia code, or an Excel Power Query, that fetches the same data straight from the source, and every result names its source with a link. reproduce_workbook returns the data as a formatted Excel workbook.

Any client that speaks MCP, including Claude, Codex, Cursor, VS Code and ChatGPT in developer mode. The Connect page has the steps.

Open an issue on GitHub with the question and the source URL from the result. The Contribute page explains what helps most.