Stop Browsing data.gouv.fr — Just Ask Your Chatbot Instead

You know the drill: you need a specific French government dataset, so you open data.gouv.fr and start clicking through categories, typing search queries, and squinting at metadata pages. It works, but it's slow. What if you could just ask your AI assistant "Quels jeux de données sont disponibles sur les prix de l'immobilier ?" and get an instant, accurate answer? That's exactly what the data.gouv.fr MCP Server is built for.

This project wraps the French national Open Data platform in a Model Context Protocol (MCP) server, letting AI chatbots like Claude, ChatGPT, and Gemini search, explore, and analyze datasets directly through conversation. No more manual browsing — just ask and get answers.

What It Does

At its core, this is an MCP server that bridges the gap between conversational AI and the data.gouv.fr platform. MCP, or Model Context Protocol, is the standard that lets AI assistants connect to external tools and data sources. This server implements that protocol specifically for data.gouv.fr, giving chatbots a structured way to query the platform's vast catalog of French public datasets.

The architecture is straightforward: you connect your preferred AI client to the server endpoint, and the server handles the communication with data.gouv.fr's API. It's built on the MCP specification, which means it works with a wide range of clients — the README lists support for ChatGPT, Claude Desktop, Claude Code, Cursor, VS Code, Gemini CLI, and more than a dozen others.

The project is MIT-licensed and maintained by the data.gouv.fr team themselves, which is a good sign for longevity. It's also continuously integrated via CircleCI, so it's not a weekend hack — this is an actively maintained piece of infrastructure.

Why It's Cool

The most striking thing about this project is how it eliminates friction. Instead of learning the quirks of a government data portal's search interface, you just ask a natural-language question. That's genuinely useful for journalists, researchers, civic hackers, or anyone who works with French public data regularly.

• Zero-config public endpoint: You don't need to host anything. The team runs a public instance at https://mcp.data.gouv.fr/mcp with no access restrictions. Just point your chatbot at it and you're done. That's a huge barrier removed for casual users.

• Broad client compatibility: The setup instructions cover 16 different clients, from consumer-facing options like ChatGPT and Le Chat to developer tools like Claude Code and Cursor. Whether you're a command-line person or a GUI enthusiast, there's a path in.

• Plain French queries: The examples in the README show the real use case — asking about property prices or population data in natural French. This isn't an English-first tool with French support bolted on; it's built for the platform's actual users.

• Self-hosting is an option: If you don't want to rely on the public instance, you can run your own and swap in your URL. The README makes this clear by noting you can "self-host" and use your own endpoint.

The design decision to use a hosted endpoint with no restrictions is particularly smart. It means the team can gather feedback and iterate while providing immediate value. They even link to a feedback form, which suggests they're actively listening to how people use it.

How to Try It

Getting started is remarkably simple, especially if you use the hosted endpoint. Here's the quickest path:

1. Pick your client. The README has detailed instructions for 16 different tools. Let's say you use Claude Code — you'd run:

claude mcp add --transport http datagouv https://mcp.data.gouv.fr/mcp

1. If you use Claude Desktop, you'd add this to your config file (typically at ~/.config/Claude/claude_desktop_config.json on Linux):

{
  "mcpServers": {
    "datagouv": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.data.gouv.fr/mcp"]
    }
  }
}

1. For ChatGPT, you'll need a paid plan (Plus, Pro, Team, or Enterprise). Then it's a few clicks through Settings > Apps and connectors, enabling Developer mode, and adding the connector URL.

2. Start asking questions. Try something like "Montre-moi les dernières données de population pour Paris" and see what comes back.

If you're using a client that isn't in the list, the pattern is the same — you're just registering an HTTP MCP endpoint. The full configuration details for every supported client are in the repository README.

Final Thoughts

This is one of those projects that feels obvious in hindsight. Government data portals are notoriously clunky to navigate, and AI assistants are getting good at natural-language queries. Bridging the two was inevitable — but data.gouv.fr actually did it, and did it well.

The hosted endpoint with no restrictions is the killer feature here. You can try it in five minutes with zero setup, which is exactly how developer tools should work. If you work with French public data in any capacity, this will save you real time. If you don't, it's still a clean example of how MCP servers can make existing platforms dramatically more accessible.

Worth a bookmark, and worth watching — the team clearly plans to iterate based on feedback.

Follow @githubprojects for more developer tools and open source projects.