Skip to content

MCP server · South Africa

Cited facts about any place in South Africa, inside your AI assistant.

GroundTruthData serves crime, rainfall, population, boundaries and mapped facilities from public sources to Claude, ChatGPT and other MCP clients. Answers are aggregated, and every number carries its source, licence and date.

Plans from R1,499 a month · 8 public sources · 2026 boundaries

Assistant session · groundtruthdata connected

Question

We're assessing a site in Stellenbosch. How many residential burglaries did the local police station record last financial year, how many people live in the municipality, and how wet was last winter?

Tools called

  • resolve_location
  • compare_areas
  • query_metrics

Answer

Stellenbosch police station recorded 612 burglaries at residential premises in FY2024/25[1]. The count is by police precinct, which does not align with ward or municipal boundaries, so it is not divided by population.

Stellenbosch Local Municipality has about 190,000 residents[2], a modelled estimate. June to August rainfall was 412 mm across the municipality[3].

Citations

  1. [1] SAPS Crime Statistics, FY2024/25 · SAPS copyright · retrieved 2026-11-02
  2. [2] WorldPop R2025A 100 m · CC BY 4.0 · summed over municipality, 2025
  3. [3] CHIRPS v3.0 monthly · CC BY 4.0 · Jun–Aug 2026 total, final
Illustrative example. The figures are not real results.

How it works

A request passes through three steps

GroundTruthData runs behind the assistant you already use, as a set of bounded tools that the assistant calls.

  1. 1

    Connection

    Add GroundTruthData as a custom connector in Claude or ChatGPT and sign in, or put an API key into Claude Code or any other MCP client.

    https://mcp.groundtruthdata.co.za/mcp
  2. 2

    Request

    The assistant takes a question such as how many people live in each ward of Stellenbosch and calls the tools it needs: resolve the place, query the metric, compare areas.

    query_metrics(metric="population_total", area=…)
  3. 3

    Response

    The tool returns aggregated numbers at the source's own geography, each with source, licence, retrieval date and dataset version. Converted figures are labelled as estimates.

    source · licence · retrieved · version

Architecture

What sits behind the tools

The assistant never reaches the datasets. It calls the MCP server, which answers from the catalogued metrics and returns citations with each result.

Request path from an AI assistant to the data sourcesAn MCP client calls GroundTruthData's MCP server. The server answers from 8 public sources through 21 catalogued metrics on 5 geographies.MCP clientClaude, ChatGPT, any MCP clienttool callGroundTruthData MCP21 metric ids · 5 geographiesaggregated, row-capped answersanswer + citations8 public sourcesRainfall1Boundaries4Land use1Crime1Population1
Figure 1. The request path, with counts taken from our own contracts. Source: sources/*/source.yaml (8 sources, grouped by domain) and catalog/metrics.yaml (21 metric ids, 5 geographies).

Coverage

Public sources, joined on one map

Each source has a written contract that records its owner, licence, attribution, native geography and refresh schedule.

DomainSourcesOwners
Rainfall1Climate Hazards Center, University of California Santa Barbara
Boundaries4Municipal Demarcation Board, South African Police Service
Land use1OpenStreetMap contributors
Crime1South African Police Service
Population1WorldPop, University of Southampton
Every source, licence and attribution

Answers

What an answer contains

GroundTruthData gives an assistant sourced numbers to quote instead of figures recalled from training data.

Properties of every answer
ProvenanceEvery number carries its source, licence, retrieval date and dataset version, so an answer can be checked and quoted.
GeographyWards, municipalities, police precincts and grids are joined on one spatial index. A figure is never converted to another geography without being labelled an estimate.
BoundsAnswers are aggregated and capped by rows and by areas per call. There are no bulk exports and no personal information.
AccessOAuth 2.1 sign-in for connectors and API keys for agents. It works with any client that supports MCP.

Pricing

Flat monthly plans with hard limits

Every plan includes every source and tool. Plans differ in how much you can ask. There are no credits and no overage: calls pause until the window resets.

PlanPer monthCalls per month
StarterR1,4995,000
ProfessionalR4,99940,000
BusinessR14,999200,000
EnterpriseAgreed by email
Plans in full, with every limit

FAQ

Common questions

What is GroundTruthData?

A paid MCP (Model Context Protocol) server. Connect it to Claude, ChatGPT or any MCP client and your assistant can answer questions about South African places with aggregated, cited facts on crime, rainfall, population, boundaries and mapped facilities.

Where do the numbers come from?

Public sources: SAPS crime statistics and police precinct boundaries, Municipal Demarcation Board 2026 boundaries, CHIRPS rainfall, WorldPop population and OpenStreetMap facility counts. Every number in an answer carries its source, licence, retrieval date and dataset version. See the coverage page for the full list.

Can I download the raw datasets?

No. We sell answers, not data. Tools return bounded, aggregated results with attribution, never bulk exports of source rows. That keeps us within the source licences and keeps answers honest.

How do I connect it?

Add https://mcp.groundtruthdata.co.za/mcp as a custom connector in Claude or ChatGPT and sign in, or use an API key from your dashboard with Claude Code or any MCP client. The quickstart walks through each one in a few minutes.

What happens when I reach my limit?

Calls are refused until the daily or monthly window resets. There are no overage charges and no surprise bills. We email you at 80% and 100% of your monthly limit, and you can upgrade at any time.

All questions

Plans start at R1,499 a month. Connecting takes a few minutes with the quickstart.