INS MCP

What it is

INS MCP is a local server implementing the Model Context Protocol (MCP), the open standard by which AI assistants call external tools. It connects an MCP-compatible client (Claude Desktop, Cursor, VS Code, and others) to Tunisia’s national statistics portal — the Institut National de la Statistique (INS) data portal at dataportal.ins.tn, built on the proprietary Prognoz platform. The portal holds dimensional time-series cubes covering national accounts, the 2014 census, population projections, employment, household living conditions, and foreign trade.

The portal’s API speaks XML in and XML out, with dirty, trap-laden metadata. INS MCP’s defining job is to own all of that: its tools take structured parameters and return clean JSON, so the AI never handles a raw XML envelope. It is built for researchers, data scientists, economists, and journalists working with Tunisian official statistics. INS MCP is an independent project; it is not endorsed by or affiliated with INS.

How it works

A user asks a question in natural language. The AI client, guided by the server’s tool descriptions, chains the tools: it searches the catalogue for the right statistical cube, drills or searches that cube’s dimensions to resolve the numeric codes for the wanted indicator and region, then requests the observations for that selection. The server builds the XML request, calls the live INS portal, parses and cleans the response, and returns tidy JSON records — each with its period, value, unit, and a mandatory source-attribution block. Identity is always handled by code, not by the portal’s dirty labels, and three distinct availability signals (declared period, has-data flag, real coverage) keep the answer honest.

Use cases

• Academic research — an economist asks for Tunisia’s population projections to 2044 by governorate and sex, and gets a tidy, source-cited series instead of writing Prognoz XML by hand.
• Investigative journalism — a journalist asks for GDP at market prices over 2014–2017, or value-added by sector, and receives exact figures with the INS source reference ready to publish.
• Public policy — an analyst queries the 2014 census (population by governorate, household equipment, employment), and the server flags which census cubes are populated versus published-but-empty, so no time is wasted on empty shells.

Demo — illustrative example

{
"source_id": "OBJ5259889",
"observations": [
  { "period": "2014", "frequency": "annual", "value": 80865.5 },
  { "period": "2015", "frequency": "annual", "value": 84689.2 },
  { "period": "2016", "frequency": "annual", "value": 89792.2 },
  { "period": "2017", "frequency": "annual", "value": 96324.7 }
],
"coverage": { "start": "2014", "end": "2017", "count": 4 },
"attribution": {
  "producer": "Institut National de la Statistique (INS), Tunisie",
  "source_url": "http://dataportal.ins.tn/",
  "license_url": "https://www.ins.tn/conditions-utilisation",
  "last_updated": "2017-02-27",
  "disclaimer": "Provided as-is; not an official INS publication and not endorsed by INS."
}
}

Status and roadmap

INS MCP is open source under the MIT License and self-hosted: there is no hosted endpoint; it runs as a local stdio process that the user’s MCP client spawns — the same model as agridata MCP. The GitHub repository is the single source for installation, code, and development updates: clone it, create a virtual environment, run pip install -e ., then python -m ins_mcp. Planned next steps include a lower-friction packaged install for non-technical users, optional offline resilience, sharper discovery (transliteration and Arabic place-name aliases), and broader end-to-end testing against the live portal.

Technical report

Citable archival version of record: Gasmi, T. (2026). ins-tunisia-mcp: An MCP Server for Conversational Access to Tunisia’s National Statistics (0.1.0). Zenodo. https://doi.org/10.5281/zenodo.20738616

Source code: github.com/tanitdata/ins-tunisia-mcp