Skip to main content

​
What are MCP clients?

MCP clients are AI applications that support the Model Context Protocol (MCP). These clients can connect to MCP servers to access additional tools and data sources during conversations. The data.gouv.fr MCP server allows you to search, explore, and analyze datasets from France’s national Open Data platform directly through AI conversations, without manually browsing the website.

​
Supported clients

The data.gouv.fr MCP server supports a wide range of AI clients:
  • Claude Desktop and Claude Code - Anthropic’s desktop and CLI clients
  • ChatGPT - OpenAI’s web-based chatbot (paid plans only)
  • Gemini CLI - Google’s command-line interface
  • Cursor - AI-powered code editor
  • VS Code - Microsoft’s code editor with MCP support
  • Windsurf - Codeium’s AI-powered editor
  • HuggingChat - Hugging Face’s web-based chatbot
  • IBM Bob - IBM’s AI assistant
  • Kiro CLI and Kiro IDE - Kiro’s command-line and IDE tools
  • Le Chat (Mistral) - Mistral’s web-based chatbot
  • Mistral Vibe CLI - Mistral’s command-line interface
  • AnythingLLM - Multi-model LLM interface
For detailed setup instructions for Gemini, Cursor, VS Code, Windsurf, and other clients, see Other AI clients.

​
Hosted endpoint vs self-hosting

The easiest way to connect is using the public hosted endpoint: 
https://mcp.data.gouv.fr/mcp
This endpoint is available to everyone with no access restrictions or API keys required. The MCP server only exposes read-only tools, so no authentication is needed.

​
Self-hosting

You can also run the MCP server locally on your own infrastructure. If you choose to self-host, simply replace https://mcp.data.gouv.fr/mcp with your own server URL in the configuration examples. For self-hosting instructions, see the Run locally guide.

​
General MCP configuration structure

Most MCP clients use JSON configuration files to connect to servers. The configuration structure varies by client, but typically includes:
  • Server name - A unique identifier for the server (e.g., "datagouv")
  • Connection method - Either HTTP URL or command-based (using npx mcp-remote)
  • Server URL - The endpoint URL (https://mcp.data.gouv.fr/mcp)
  • Transport type - Usually http, streamable, or streamable-http

​
HTTP transport

The data.gouv.fr MCP server uses Streamable HTTP transport only. STDIO and SSE transports are not supported. Some clients connect directly via HTTP URL: 
{
  "mcpServers": {
    "datagouv": {
      "url": "https://mcp.data.gouv.fr/mcp",
      "type": "streamable-http"
    }
  }
}

​
Command-based transport

Other clients use npx mcp-remote as a bridge: 
{
  "mcpServers": {
    "datagouv": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.data.gouv.fr/mcp"
      ]
    }
  }
}

​
Next steps

Choose your client and follow the setup guide: Once connected, you can start asking questions like:
  • “Quels jeux de données sont disponibles sur les prix de l’immobilier ?”
  • “Montre-moi les dernières données de population pour Paris”
  • “Search for datasets about public transportation in Lyon”

Build docs developers (and LLMs) love

Get started for freeTalk to us