Most open-data portals assume a human will click through search filters, preview tables, and read metadata pages one at a time. This flips that around: the LLM does the browsing, so you ask "which datasets cover Paris real-estate prices?" in plain language and get answers pulled live from data.gouv.fr's own Main, Tabular, and Metrics APIs — no scraping, no stale CSV exports.
What Sets It Apart
- Official and hosted: maintained by the data.gouv.fr team itself, with a public Streamable HTTP endpoint (mcp.data.gouv.fr) any MCP client can point at with zero setup or API key.
- More than dataset search: separate tools cover static datasets, the cataloged third-party APIs (dataservices, e.g. the Adresse and Sirene APIs), and usage metrics — so an agent can move from discovery to actually querying data.
- Tabular queries, not just links: query_resource_data pulls real rows from CSV/XLS resources through the Tabular API with pagination, so the model can compute over the data instead of citing a file URL.
- Read-only by design: no write tools and no key required, which keeps it safe to expose to any chatbot.
Who It's For
Great fit if you work with French public data — journalists, analysts, GovTech builders — and want a chatbot that finds and reads datasets conversationally. Look elsewhere if you need international open data, write access, or heavy analytics over very large files: the Tabular API enforces size limits (CSV ≤ 100 MB, XLSX ≤ 12.5 MB) and pagination, so it's built for exploration and targeted queries rather than bulk data engineering.
