ISTAT data for your agent

Overview

This MCP (Model Context Protocol) server gives a compatible client access to statistical data from ISTAT, the Italian national statistics institute, through its SDMX REST API. It exposes eight tools: dataset discovery, structure, constraints, codelist descriptions, concepts, data in TSV format, cache diagnostics and territorial codes.

The server keeps a memory and disk cache, limits calls to four per minute with a queue, and retries requests after a transient error. The repository also contains a skill that guides the model step by step, as well as an HTTP mode with a Dockerfile and a Helm chart.

Highlights

• Eight MCP tools: dataflow discovery, structure, constraints, codelist descriptions, concepts, data, cache diagnostics and territorial codes (REF_AREA).
• Recommended three-step workflow: discover_dataflows, then get_constraints, then get_data with dimension filters.
• Two-layer cache (memory and disk, configurable lifetimes) and a limit of four calls per minute to the ISTAT API.
• A SKILL.md skill is included in skills/istat-mcp; the README recommends installing it to reduce errors and unnecessary calls.
• Documented setup for Claude Desktop, Claude Code, Gemini CLI, VS Code and Codex CLI, plus an optional HTTP mode with Docker and Helm.

Who it is for

For anyone who wants an AI assistant to answer from statistics published by ISTAT (the README mentions population and agriculture, for example) without calling the SDMX API by hand. Basic technical skills are needed (Python, configuring an MCP client).

Installation

Clone the repository, create the environment with uv (or pip; Python 3.11 or later), then declare the server in the MCP client configuration, pointing to the interpreter in the .venv folder. The README also recommends installing the bundled skill.

git clone https://github.com/ondata/istat_mcp_server.git

uv sync

claude mcp add -s user istat -- "/path/to/istat_mcp_server/.venv/bin/python" -m istat_mcp_server

Read the README on GitHub

Points of attention

• The README states there are no versioned releases: whatever is merged into main is what users install.
• The server makes at most four calls per minute to the ISTAT API, with queuing; some constraint responses can be slow (configurable timeout).
• The configuration examples must use the interpreter from the .venv folder created at install time; a generic python3 from the PATH usually cannot import the package.

Can an agent use it?

Agent-ready

The repository is an MCP server and ships a SKILL.md skill in skills/istat-mcp. The README gives configuration examples for Claude Desktop, Claude Code, Gemini CLI, VS Code and Codex CLI. There is no AGENTS.md or CLAUDE.md at the root.