One MCP to rule them all.

The other seven

• Statisticians A bachelor's degree, province by province
• Demographers Who comes to stay: Edmonton and Calgary
• Urban planners Canada is building up, not out
• Microeconomists Low income across immigrant generations
• Marketers What a rewards card costs
• Scientists Machine learning at the patent office
• Analysts The policy rate since 2015

30

federal and national sources

56

provincial, territorial and city sources

12/13

provinces and territories with local data

215

tools behind one connection

Statistics Canada, in depth Every table, series and microdata file an agent can reach, and the hard parts MapleStats handles for you. →

See every source and where it coversBrowse all 215 tools

1. 01

   Planplan_query

   The planner reads the question and lists the agencies to ask, in order, with the caveats on combining them. Places it recognizes add their local portals.

   Question

   How have rents and interest rates moved in Calgary since 2020?

   Plan

   Housing: starts, rents, prices, mortgages

   1. cmhc_list_categoriesCMHC starts, completions, rents and vacancy by centre
   2. cmhc_get_table_datapull the CMHC table for the place and period
   3. wds_search_cubesStatCan New Housing Price Index, building permits
   4. boc_search_seriesBank of Canada mortgage and policy rates
   5. statcan_census_profile_get_datacensus shelter cost and tenure by area

   Caveat CMHC reports by census metropolitan area and centre; StatCan price indexes are by CMA too, but the two define some areas differently, so name the geography each figure uses.

   Interest rates, exchange rates and markets

   1. boc_search_seriesfind the Valet series (e.g. FXUSDCAD, V39079)
   2. boc_get_observationspull the series for the period

   Calgary (city)

   1. socrata_search_datasetssearch with portal='calgary'

2. 02

   Findsearch_tools

   Search ranks the catalogue against a plain-language query and returns the top 5. French works because every tool carries French keywords. The panel runs the server's own index in your browser, so type your own query.

   search_toolsRuns in your browser

   Describe the data you need

   >

   Try federal contract awards PUMF bootstrap weights tide times Halifax census profile income taux de chômage

   1. 1boc_search_seriesBank of CanadaSearch Bank of Canada Valet's ~16,000 statistical series by keyword.
   2. 2boc_get_seriesBank of CanadaGet the label and description for one Bank of Canada Valet series.
   3. 3boc_get_observationsBank of CanadaGet observations for one or more Bank of Canada Valet series.
   4. 4boc_get_groupBank of CanadaGet a Bank of Canada Valet group's description and member series.
   5. 5boc_search_groupsBank of CanadaSearch Bank of Canada Valet's series groups by keyword.

   Top 5 of 214 searchable tools, ranked by the same BM25 index search_tools uses.

3. 03

   Fetchcall_tool

   The agent calls the tool by name. The result is a typed object, and its provenance block, marked here, gives the exact upstream URL and the time of the query.

   Request

   {"name": "boc_get_observations", "arguments": {"series_names": ["V39079"], "recent": 3}}

   Result

   Target for the overnight rate (business daily) V39079

   Date	Value
   24 September 2026	2.25
   23 September 2026	2.25
   22 September 2026	2.25

   Provenance

   Source URL

   https://www.bankofcanada.ca/valet/observations/V39079/json

   Queried

   27 September 2026, 17:00 UTC

   Result type

   boc.ObservationsResult

   Show the full response as JSON

   {
     "series": {
       "V39079": {"name": "V39079", "label": "Target for the overnight rate (business daily)", "description": "Also called the policy interest rate, the average rate that the Bank of Canada wants to see in the market for overnight money market financing. (V39079)"}
     },
     "observations": [
       {"ref_date": "2026-09-24", "values": {"V39079": 2.25}},
       {"ref_date": "2026-09-23", "values": {"V39079": 2.25}},
       {"ref_date": "2026-09-22", "values": {"V39079": 2.25}}
     ],
     "provenance": {
       "source": "boc",
       "url": "https://www.bankofcanada.ca/valet/observations/V39079/json",
       "queried_at": "2026-09-27T17:00:48.227537Z",
       "as_of": null,
       "freshness": null,
       "coverage": null,
       "limits": null,
       "cached": false,
       "schema_name": "boc.ObservationsResult",
       "reproduce": "For R, Python, Stata and Julia scripts that fetch and clean this data, call reproduce_code with this tool's name and arguments."
     }}

4. 04

   Reproducereproduce_code

   The same call becomes an R, Python, Stata or Julia script that downloads, checks and cleans the data, so the analysis does not depend on the agent.

   Request

   {"tool_name": "boc_get_observations", "arguments": {"series_names": ["V39079"], "recent": 3}}

   RPythonStataJulia

   R

   # ============================================================
   # Bank of Canada Valet: V39079
   # Purpose: Fetch the data behind MapleStats MCP's boc_get_observations
   #          (exact: Valet request rebuilt from the tool's arguments)
   # Inputs:  https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3
   # Outputs: data/raw/valet_observations.json; the prepared table as `data`
   # ============================================================
   
   # 0. Setup ----
   
   library(dplyr)
   library(httr2)
   library(janitor)
   library(jsonlite)
   library(lubridate)
   library(readr)
   library(stringr)
   library(tibble)
   library(tidyr)
   
   dir.create("data/raw", recursive = TRUE, showWarnings = FALSE)
   
   # 1. Read inputs ----
   
   response <- request("https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3") |>
     req_perform()
   
   writeLines(resp_body_string(response), "data/raw/valet_observations.json")
   
   payload <- fromJSON("data/raw/valet_observations.json", flatten = TRUE)
   data <- as_tibble(payload[["observations"]])
   
   # 2. Check inputs ----
   
   stopifnot("https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3 returned no rows" = nrow(data) > 0)
   
   # 3. Prepare data ----
   
   data <- data |>
     clean_names()
   
   # Valet nests each series as <series>.v; make one row per date and series.
   
   data <- data |>
     pivot_longer(-d, names_to = "series", values_to = "value") |>
     mutate(
       series = str_remove(series, "_v$") |> str_to_upper(),
       value = as.numeric(value),
       date = ymd(d)
     ) |>
     select(date, series, value)
   
   # Standard cleaning: trimmed text, empty strings as missing, and numbers
   # stored as text converted to numbers.
   
   data <- data |>
     mutate(across(where(is.character), \(x) na_if(str_trim(x), ""))) |>
     type_convert()

   Python

   # ============================================================
   # Bank of Canada Valet: V39079
   # Purpose: Fetch the data behind MapleStats MCP's boc_get_observations
   #          (exact: Valet request rebuilt from the tool's arguments)
   # Inputs:  https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3
   # Outputs: data/raw/valet_observations.json; the prepared table as `data`
   # ============================================================
   
   # %% 0. Setup
   
   import json
   import re
   import unicodedata
   from pathlib import Path
   
   import httpx
   import polars as pl
   
   RAW_DIR = Path("data/raw")
   RAW_DIR.mkdir(parents=True, exist_ok=True)
   raw_path = RAW_DIR / "valet_observations.json"
   
   # %% 1. Read inputs
   
   with httpx.Client(
       http2=True,
       follow_redirects=True,
       timeout=300,
       headers={"User-Agent": "Mozilla/5.0 (compatible; research script)"},
   ) as client:
       response = client.get('https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3')
   response.raise_for_status()
   raw_path.write_bytes(response.content)
   
   payload = json.loads(raw_path.read_text(encoding="utf-8"))
   records = payload['observations']
   data = pl.json_normalize(records, strict=False)
   
   # %% 2. Check inputs
   
   assert data.height > 0, "https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3 returned no rows"
   
   # %% 3. Prepare data
   
   # Valet nests each series as <series>.v; make one row per date and series.
   
   data = data.unpivot(index="d", variable_name="series", value_name="value")
   data = data.with_columns(
       pl.col("d").str.to_date().alias("date"),
       pl.col("series").str.replace(r"\.v$", ""),
       pl.col("value").cast(pl.Float64, strict=False),
   ).select("date", "series", "value")
   
   # Standard cleaning: snake_case names without accents (PÉRIODE -> periode,
   # referenceNumber -> reference_number, as janitor does in R), trimmed text,
   # empty strings as missing. Names that clean alike are numbered as janitor
   # numbers them (Indicator, indicator -> indicator, indicator_2).
   
   clean_names = [
       re.sub(
           r"[^0-9a-z]+",
           "_",
           re.sub(
               r"([a-z0-9])([A-Z])",
               r"\1_\2",
               unicodedata.normalize("NFKD", column).encode("ascii", "ignore").decode(),
           ).lower(),
       ).strip("_")
       for column in data.columns
   ]
   while len(set(clean_names)) < len(clean_names):
       name_counts = {}
       numbered = []
       for name in clean_names:
           name_counts[name] = name_counts.get(name, 0) + 1
           count = name_counts[name]
           numbered.append(name if count == 1 else f"{name}_{count}")
       clean_names = numbered
   data = data.rename(dict(zip(data.columns, clean_names)))
   data = data.with_columns(pl.col(pl.Utf8).str.strip_chars().replace("", None))

   Stata

   * ============================================================
   * Bank of Canada Valet: V39079
   * Purpose: Fetch the data behind MapleStats MCP's boc_get_observations
   *          (exact: Valet request rebuilt from the tool's arguments)
   * Inputs:  https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3
   * Outputs: data/raw/valet_observations.json; the prepared table as `data`
   * ============================================================
   
   version 18
   clear all
   set more off
   
   * 0. Setup
   
   capture mkdir "logs"
   capture log close
   log using "logs/boc_get_observations.log", replace
   capture mkdir "data"
   capture mkdir "data/raw"
   
   * 1. Read inputs
   
   * Stata reads no JSON or HTML and truncates long column names, so its
   * built-in Python (Stata 16+) fetches, filters and writes a CSV. Point
   * Stata at a Python with these packages first: python set exec <path>.
   
   python:
   import json
   from pathlib import Path
   import httpx
   import polars as pl
   RAW_DIR = Path("data/raw")
   RAW_DIR.mkdir(parents=True, exist_ok=True)
   raw_path = RAW_DIR / "valet_observations.json"
   with httpx.Client(http2=True, follow_redirects=True, timeout=300, headers={"User-Agent": "Mozilla/5.0 (compatible; research script)"}) as client: response = client.get('https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3')
   response.raise_for_status()
   raw_path.write_bytes(response.content)
   payload = json.loads(raw_path.read_text(encoding="utf-8"))
   records = payload['observations']
   data = pl.json_normalize(records, strict=False)
   nested = [name for name, dtype in data.schema.items() if dtype.is_nested()]
   data = data.with_columns(pl.col(name).map_elements(lambda value: json.dumps(value.to_list() if isinstance(value, pl.Series) else value, default=str), return_dtype=pl.Utf8) for name in nested)
   data.write_csv(RAW_DIR / "valet_observations_prepared.csv")
   end
   
   import delimited "data/raw/valet_observations_prepared.csv", clear varnames(1) encoding("utf-8")
   
   * 2. Check inputs
   
   assert _N > 0
   
   * 3. Prepare data
   
   * One column per series (<series>_v), one row per date.
   
   generate date = date(d, "YMD")
   format date %td
   drop d
   
   * Standard cleaning: lower-case names, trimmed text, and numbers stored as
   * text converted (destring leaves genuinely non-numeric text alone).
   * rename *, lower stops at a clash (Indicator next to indicator), so names
   * that lower-case alike are numbered as janitor numbers them (indicator,
   * indicator_2), then renamed in one group rename, which allows swaps.
   
   local names
   foreach var of varlist _all {
       local names `names' `=strlower("`var'")'
   }
   local dups : list dups names
   while "`dups'" != "" {
       local numbered
       local before
       foreach name of local names {
           local count 1
           foreach earlier of local before {
               if "`earlier'" == "`name'" local ++count
           }
           local before `before' `name'
           if `count' > 1 {
               local name = substr("`name'", 1, 32 - strlen("_`count'")) + "_`count'"
           }
           local numbered `numbered' `name'
       }
       local names `numbered'
       local dups : list dups names
   }
   local old_names
   local new_names
   local i 0
   foreach var of varlist _all {
       local ++i
       local name : word `i' of `names'
       if "`name'" != "`var'" {
           local old_names `old_names' `var'
           local new_names `new_names' `name'
       }
   }
   if "`old_names'" != "" {
       rename (`old_names') (`new_names')
   }
   quietly ds, has(type string)
   local text_vars `r(varlist)'
   foreach var of local text_vars {
       replace `var' = strtrim(`var')
   }
   destring, replace
   
   log close

   Julia

   # ============================================================
   # Bank of Canada Valet: V39079
   # Purpose: Fetch the data behind MapleStats MCP's boc_get_observations
   #          (exact: Valet request rebuilt from the tool's arguments)
   # Inputs:  https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3
   # Outputs: data/raw/valet_observations.json; the prepared table as `data`
   # ============================================================
   
   # 0. Setup
   
   using DataFrames
   using Downloads
   using JSON3
   using Tables
   using TidierData
   
   mkpath("data/raw")
   
   # 1. Read inputs
   
   Downloads.download("https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3", "data/raw/valet_observations.json")
   payload = JSON3.read(read("data/raw/valet_observations.json", String))
   records = payload["observations"]
   data = DataFrame(Tables.dictrowtable(records))
   
   # 2. Check inputs
   
   @assert nrow(data) > 0 "https://www.bankofcanada.ca/valet/observations/V39079/json?recent=3 returned no rows"
   
   # 3. Prepare data
   
   # Standard cleaning: snake_case names, trimmed text, empty strings as missing.
   
   data = @chain data begin
       @clean_names
   end
   data = mapcols(
       col -> eltype(col) <: Union{Missing, AbstractString} ?
           [ismissing(x) || isempty(strip(x)) ? missing : strip(x) for x in col] : col,
       data,
   )