Community Pulseby Data n Dashboards

Documentation

Community Pulse, from link to first answer

Ask plain-English questions about New Zealand regional data from inside Claude, and get answers backed by real query results. This page covers setting the connector up, what it can and can't answer, and what to do when something looks off.

Add the connector Ask your first question Troubleshooting
On this page

On this page

  1. How it works
  2. Add the connector
  3. Your first question
  4. What you can ask
  5. Ready-made reports
  6. The tools Claude uses
  7. How answers stay honest
  8. Accounts, plans and limits
  9. Troubleshooting
  10. Get help

How it works #

You ask. Claude works out which tools to call. The connector runs a structured query and hands back the numbers, plus everything you'd need to check them.

"How many people work in manufacturing in the Waikato region?"
  → Claude calls list_datasets, search_values, query_dataset
  → 24,800 employees   [Geographic Units by Industry and Statistical Area · Waikato Region · Manufacturing · 2025]

Every figure Claude states comes from a real query against our data. This connector writes no prose and does no reasoning about your question — Claude does that. Its job is to expose good tools and honest data.

Matched, not guessed

Your wording is resolved against values that actually exist in the data. An ambiguous match is refused and both options offered, never silently picked.

Checkable afterwards

Each answer carries the query that ran, the unit, the resolved place and period, and any caveat — so a number you quote in public can be traced.

Read-only, always

There is no free-text query tool and no write path back to the source — not behind a flag, not for an administrator.

Add the connector #

About a minute, done once per person. Everyone on your team signs in with their own account, so this isn't something one person sets up on behalf of everybody else.

Watch first, if you prefer. The launch video walks through adding the connector and asking a question. The steps below cover the same ground in writing.

Paste it exactly as shown — nothing before or after /mcp. A trailing slash or an extra path segment is the most common reason a connector won't add.

  1. Open Claude's connector settings

    In Claude, go to Settings, then Connectors, then Add custom connector.

    Settings › Connectors › Add custom connector

  2. Paste the link and name it

    “Community Pulse” is fine. Leave every OAuth field (Client ID, Client Secret) blank — Claude discovers and registers the connection itself, so there is nothing correct to put in them.

  3. Turn it on for your chat

    Depending on your Claude client, a new connector is either enabled automatically or needs switching on from the + / tools menu inside the conversation. Check there first if a question doesn't seem to reach it.

  4. Sign in when prompted

    The first real tool call opens this site's own sign-in page, separate from your Claude account. Create an account or log in — new accounts start on the Free plan, no card required.

  5. Ask a real question

    Try one of the examples below, or ask about your own district. A working answer names its source dataset and period, like [Geographic Units by Industry and Statistical Area · Waikato Region · 2025].

Adding a custom connector needs a Claude plan that allows it. If Add custom connector isn't in your Claude settings at all, that's a Claude account setting rather than anything on this end — check with whoever administers your Claude workspace.

Your first question #

No special syntax, no official field names. Click a question to copy it, then paste it into Claude.

What a good answer looks like

A figure on its own isn't an answer you can put in a report. Every query_dataset result comes back with the dataset it came from, the unit, the resolved place and period, and any caveat that applies — a filter value that was snapped to the nearest real one, a partial year that was dropped, a grand-total row that was pinned. Claude should be passing those on to you; if a number arrives bare, ask it which dataset and year the figure came from.

Ask “What can I ask?” in Claude at any time for a live tour of what your plan covers — it reads the real catalog rather than this page, so it is never out of date.

What you can ask #

New Zealand regional statistics, drawn from the wider Data n Dashboards data lake across these national themes.

  • Population & Demographics
  • Economy & National Accounts
  • Labour Market & Business
  • Income, Living Standards & Equity
  • Housing & Built Environment
  • Health & Wellbeing
  • Education & Skills
  • Transport & Infrastructure
  • Environment, Climate & Energy
  • Society & Communities
  • Tourism & International Travel
  • Governance, Elections & Public Sector

Live today through this connector, for example: Geographic Units by Industry and Statistical Area (employee and business-location counts, by industry and area), Gross Domestic Product (by industry and region or district), Road Accidents (crashes by severity), Victimisations Time and Place (recorded crime, monthly), subnational population estimates, average house prices, inflation, natural disaster cost and air quality. list_datasets always has the live, exact list for your plan — treat that, not this paragraph, as the source of truth.

Geography

Most datasets can be filtered at region and territorial authority (city or district) level, and some down to ward or SA2 (suburb). Ask at one level at a time: filtering a region and one of its own districts in the same question would double-count, so the connector keeps the finest level given and tells you what it dropped. list_areas tells Claude which districts sit inside which region.

What it won't do

Rates it can't support

Not every dataset carries population, so a per-capita rate is only produced when the population dataset is paired in deliberately, as a second query, and labelled as its own arithmetic.

Periods with no data

A year outside a dataset's coverage is refused rather than answered as zero. On a row-counting dataset, “0” would be a real-looking figure stating something false.

Cause and effect

There is no control group in this data. “Why did it move?” can be answered by decomposition and known measurement artefacts — not by proving a cause.

Anything personal

These are published aggregate statistics. There are no records about identifiable people to ask for.

Ready-made reports #

From the + menu in a Claude chat you can pick a pre-built analysis instead of asking freeform. You choose these — Claude doesn't pick them for you.

Regional profile

Jobs, GDP, road crashes and crime for one place, in one document.

regional_profile

Economy snapshot

Employment and GDP together, and a plain read on where the two don't agree.

economy_snapshot

Safety snapshot

Crime and road crashes side by side — counts only, never a rate that isn't there.

safety_snapshot

Industry deep dive

One industry across jobs and GDP — its size, its rank, its trend.

industry_deep_dive

Compare areas

Two places, checked first that they're at the same geographic level.

compare_areas

Trend report

Change over time, with partial years and month ordering already handled.

trend_report

Policy brief

A leadership-ready briefing. It asks who it's for and what scope before writing a word.

policy_brief

Explain why

Investigates a change, ranking what the data confirms above what would only be speculation.

explain_why

Infographic

One self-contained HTML page built from real tool calls — and no invented imagery of a real place.

infographic

Motion graphic

A short animated explainer video on any topic or place, made with Claude Motion — real numbers, and motion that never exaggerates a change.

motion_graphic

What can I ask

A guided tour of what the data covers, and what it plainly doesn't.

what_can_i_ask

The tools Claude uses #

Seven purpose-built tools. Only one of them ever returns a number, and there is no free-text query tool at all — Claude never writes SQL.

list_datasets

What data is available on your plan right now.

describe_dataset

The exact columns, units and usage rules for one dataset.

search_metadata

Which dataset covers a concept, when the name isn't obvious.

search_values

Resolves a place or category name to its exact value in the data.

query_datasetReturns numbers

The only tool that returns figures — always an aggregate, never raw rows. Carries the query, the unit, the resolved scope and every caveat with each result.

get_geojson

Boundary polygons for drawing a map. Geometry only, never data values.

list_areas

The Region / territorial authority hierarchy — which districts sit inside which region.

For your IT team: the connector speaks the Model Context Protocol over Streamable HTTP with JSON responses. Every tool is read-only, tool arguments are structured rather than free text, and the source system is only ever read from.

How answers stay honest #

Most of the work in this connector goes into refusing to hand back a number it can't stand behind.

Refusing beats guessing

A result whose filters can't be proven to have applied is withheld, not returned with a warning attached. In practice the figure gets quoted and the warning doesn't.

Ambiguity is surfaced

“Waikato” matches both Waikato Region and Waikato District — different places with very different numbers. Two candidates means refuse and offer both, never pick one.

Totals don't masquerade as parts

Grand-total rows are handled in one place: pinned for an overall figure, excluded from a breakdown. The classic failure — a real-looking number that is actually the grand total — is checked for explicitly.

Time anchors on the data

“The last six months” is counted from the data's own latest period, not today's date. These extracts lag reality, and counting from the wall clock would answer “no data” to a question the data answers fine.

One thing worth knowing. A breakdown with no year given aggregates every year in the dataset, not just the latest. That's legitimate and it's always stated in the caveats — but it's rarely what someone meant. If a breakdown looks far too big, check the period first.

Usage is logged only as far as which tool ran and which dataset it touched — never the words of your question or Claude's answer. Full detail in the privacy policy.

Accounts, plans and limits #

One account per person, and a rolling call allowance rather than a charge per question.

Everyone signs in individually

By design, so every query is attributable to a real account rather than a password passed around the office. Adding the connector is a per-person step.

Calls, not questions

One question usually costs several tool calls, because Claude checks the catalog and resolves names before it queries. Both plans have a limit on tool calls within a rolling window.

Free to start

New accounts start on the Free plan with a starter set of datasets and no card required. See Pricing for the current plans and allowances.

Signing in again

Expected after a few hours, or in a new browser. Sign in the same way you did the first time — nothing needs re-adding in Claude.

Troubleshooting #

The things that actually go wrong, and what to do about each.

The connector won't add, or Claude says it can't find the tool

Almost always the URL. Re-check the pasted link ends in /mcp with nothing else appended — no trailing slash, no extra path. Copy it again from above, then remove and re-add the connector in Claude.

Leave the OAuth fields blank. Filling in a Client ID or Secret stops registration working, because there is no pre-registered client to match.

Nothing happens when I ask a question

Open the connector's entry in Claude's settings and confirm it is both connected and switched on for the chat you're in — some clients scope “on” per conversation rather than account-wide. It's usually under the + / tools menu inside the conversation.

If it's on and still quiet, ask directly: “Use the Community Pulse connector to list the datasets you can see.”

I'm asked to sign in again

Expected after a few hours, or in a new browser. Sign in the same way as the first time — you don't need to remove or re-add the connector.

Claude says a figure isn't available

That's usually the connector correctly refusing a question the data can't answer — a period outside a dataset's coverage, a rate the dataset can't support, or a place name that matched two different areas. The refusal names the tool that would fix it, so a good next move is to ask Claude to follow that suggestion.

Asking “what years does that dataset cover?” or “which areas match Waikato?” usually resolves it in one turn.

A number looks wrong, or far too big

Check the period first. A breakdown with no year given sums every year in the dataset — stated in the caveats, but easy to skim past. Ask Claude to repeat the query for a single year.

Then check the geographic level: a region and one of its districts are different rows, and some datasets hold national, regional and district rows in the same column. Ask which level the figure is at.

If it still looks wrong, tell us and include the question you asked. A wrong number is a bug we want to hear about.

I hit a call limit mid-conversation

Each plan has a rolling tool-call limit rather than a per-question charge. See Pricing for your plan's allowance, or to upgrade. A ready-made report tends to cost fewer calls than many follow-up questions covering the same ground.

Get help #