data.gouv.fr launches its MCP server: when AI talks to French public data
What if, instead of browsing a website to find air quality statistics or real estate prices, you could simply ask an AI? That's exactly what data.gouv.fr's latest experiment offers.
On February 25, 2026, France's national open data platform launched its MCP server — a gateway that allows AI chatbots like Claude, ChatGPT, and Gemini to directly access over 74,000 public datasets. A world first for a national open data platform.
What is MCP? The "USB-C" of artificial intelligence
Before diving deeper, let's clarify this acronym you'll be seeing more and more. MCP stands for Model Context Protocol. Behind this somewhat technical abbreviation lies a simple idea.
Today, AI tools like Claude, ChatGPT, and Gemini are excellent at understanding and generating text. But they have a problem: they're isolated. Their knowledge stops at their training date, and they can't, on their own, fetch information from a database or interact with software.
To solve this, developers previously had to create a custom connector for each tool: one for Google Drive, another for Slack, another for a database. With dozens of different AIs and thousands of possible tools, the equation quickly became unmanageable — this is called the N×M problem.
MCP changes the game. It's a universal protocol — a common language — that allows any AI to connect to any service, as long as that service exposes an "MCP server." The most fitting analogy: MCP is to AI what USB-C is to electronic devices. Before USB-C, every manufacturer had its own cable. Now, one cable works everywhere.
| Before MCP | With MCP |
|---|---|
| Each AI needs a specific connector for each tool | One standard protocol connects everything |
| Developers multiply custom integrations | One MCP server works with all AIs |
| Accessing data requires technical code | AI queries data in natural language |
| Each update breaks existing connectors | The standard protocol ensures compatibility |
Created by Anthropic (the company behind Claude) in late 2024, the protocol was handed to the Linux Foundation in December 2025, with the creation of the Agentic AI Foundation — co-founded with OpenAI and backed by Google and Microsoft. A strong signal: MCP has become the industry standard.
data.gouv.fr launches its MCP server: what you need to know
74,000 datasets accessible in natural language
data.gouv.fr is France's national open data platform, managed by Etalab (part of DINUM). Ranked #1 globally by the OECD for open government data, it hosts over 74,000 datasets covering a wide range of domains:
- Businesses — Sirene registry (all French companies)
- Real estate — Property transaction values
- Demographics — Deceased persons registry, population data
- Transportation — Schedules, lines, ridership
- Healthcare — Hospital statistics, epidemiology
- Environment — Air quality, weather data, biodiversity
- Elections — Results, turnout, electoral districts
Until now, using this data meant browsing the website, downloading CSV or JSON files, and often writing code to analyze them. With the MCP server, you simply ask a question in natural language from a compatible chatbot.
Available tools
The server exposes several "tools" — structured capabilities that AIs can use automatically:
| Tool | What it does |
|---|---|
| search_datasets | Search across 74,000+ datasets by keywords |
| get_dataset_info | View detailed metadata for a dataset |
| list_dataset_resources | List associated files and resources |
| query_resource_data | Directly query a resource's data |
| download_and_parse_resource | Download and analyze a complete file |
| get_metrics | Access usage and popularity indicators |
| search_by_geographic_area | Filter by geographic zone (municipalities, departments, regions) |
What can you actually do with it?
Here are concrete examples of what you can do by connecting your AI to the data.gouv.fr MCP server:
For an investigative journalist
"What are the top 10 departments by real estate transactions in 2025? Compare with 2020." — The AI queries property data, aggregates results, and produces a comparative table in seconds. Work that would have taken hours of manual research.
For a data analyst
"Find datasets about air quality in the Paris region and show me the latest available measurements." — No more manually browsing the catalog. The AI searches, identifies the right resources, and extracts relevant data.
For a curious citizen
"How many businesses were created in my city last year?" — Instant access to Sirene data, no technical skills required. This is the MCP promise: democratizing access to public data.
For a developer
"List datasets that follow the GTFS schema for public transportation." — Quick identification of standardized datasets to integrate into an application.
Why this is a turning point for open data
First government MCP server in the world
The data.gouv.fr MCP server is the first official MCP server deployed by a government to make its open data AI-accessible. Other initiatives exist — the UK offers an MCP for parliamentary debates and another for legislation — but none covers a national open data catalog of this scale.
French consulting firm OCTO calls this development the "missing link" of the Platform State concept: the vision, theorized in 2013 by Henri Verdier, of an open government whose data and services would be truly accessible to everyone — not just developers, but every citizen.
From technical APIs to natural language
Public data was already accessible via APIs. But those APIs were designed for developers: you needed to write code, understand technical documentation, and manage authentication. MCP turns every API into a conversational interface. Anyone can query public data, as long as they have access to a compatible chatbot.
Compatible with major AI chatbots
| Tool | Publisher | MCP Support |
|---|---|---|
| Claude | Anthropic | Native (protocol creator) |
| ChatGPT | OpenAI | Supported |
| Gemini | Supported | |
| Mistral | Mistral AI | Supported |
| Cursor | Cursor Inc. | Supported (AI IDE) |
| Windsurf | Codeium | Supported (AI IDE) |
Limitations to be aware of
The data.gouv.fr team is transparent: this is a strictly experimental framework. Here are the key concerns:
- AIs can make mistakes — The underlying data is official, but the AI interpreting it may produce incomplete, approximate, or incorrect responses
- Coverage is partial — Only about ten tools are exposed, while the data.gouv.fr API offers many more capabilities
- Read-only access — No modification, deletion, or addition of data is possible via MCP
- No rate limit information — The server is public and free, but no restrictions on volume have been communicated
How to try it right now
Two options are available, depending on your profile:
Option 1: Via the public endpoint (easiest)
Add the MCP server to your compatible chatbot's configuration. No API key needed:
{
"mcpServers": {
"datagouv": {
"url": "https://mcp.data.gouv.fr/mcp"
}
}
}Option 2: Local installation (for developers)
For those who want to contribute or customize the server:
git clone https://github.com/datagouv/datagouv-mcp.git
cd datagouv-mcp
uv run datagouv-mcpThe project is built with Python, under the MIT license, and contributions are welcome on GitHub.
What comes next?
This experiment opens ambitious perspectives. The data.gouv.fr team envisions eventually enabling data publishing and editing via MCP, particularly with sovereign AI models.
More broadly, consulting firm OCTO envisions the creation of 10 to 15 "foundational MCP servers" covering 80% of citizen needs: identity, address, businesses, employment, education, healthcare, taxation... An ecosystem where every public service could be queried in natural language by any AI.
We're not there yet. But with this first building block, France is laying the foundations for truly universal access to public data — beyond developers, beyond experts, open to everyone.
FAQ
Conclusion
The data.gouv.fr MCP server is more than a technical curiosity. It's the signal that access to public data is undergoing a paradigm shift. No more CSV downloads and API calls reserved for developers — tomorrow, anyone will be able to query government data by simply asking their AI a question.
The experiment is still young and its limitations are real. But the direction is clear: open data and AI are converging, and France is leading the way.
Compare MCP-compatible chatbots
Claude, ChatGPT, Gemini, Mistral — find out which AI chatbot best fits your needs, with detailed feature and pricing comparisons.
View comparisonStay up to date with the latest AI news
Get the latest artificial intelligence news delivered to your inbox every week.
No spam. Unsubscribe in 1 click.


