Skip to main content
The data.gouv.fr MCP server provides 10 tools that enable AI chatbots to search, explore, and analyze datasets from data.gouv.fr, the French national Open Data platform.
These tools are called by AI chatbots through the Model Context Protocol (MCP), not directly by users. The chatbot interprets your natural language requests and calls the appropriate tools to fulfill them.

​
Tool categories

The tools are organized into three categories:

​
Datasets

Tools for working with static data files:

​
Dataservices

Tools for working with external third-party APIs registered in the data.gouv.fr catalog:

​
Metrics

Tools for accessing usage statistics:
  • get_metrics - Get visit and download statistics for datasets and resources

​
Common workflows

​
Exploring datasets

The typical workflow for finding and analyzing dataset data:
  1. Search - Use search_datasets to find datasets matching your keywords
  2. Explore - Use list_dataset_resources to see available files in a dataset
  3. Preview - Use query_resource_data (for CSV/XLSX) or download_and_parse_resource (for JSON/JSONL) to preview the data structure
  4. Analyze - Query or download the full data as needed

​
Working with APIs

The typical workflow for discovering and using external APIs:
  1. Search - Use search_dataservices to find APIs matching your keywords
  2. Get details - Use get_dataservice_info to get metadata and documentation URLs
  3. Understand the API - Use get_dataservice_openapi_spec to see available endpoints and parameters
  4. Call the API - Use the base_api_url and endpoint information from the spec

​
Search query tips

Both search_datasets and search_dataservices use AND logic for multi-word queries:
  • Use short, specific keywords for best results
  • Generic words like “données”, “fichier”, “csv” are automatically filtered out
  • If a query returns no results, try removing generic terms or using more specific keywords

​
Data access strategies

The MCP server provides two ways to access tabular data:

​
Tabular API (query_resource_data)

  • Best for: CSV and XLSX files, exploratory analysis, filtered queries
  • Advantages: Fast, no download required, supports filtering and sorting
  • Limitations: CSV files ≤ 100 MB, XLSX files ≤ 12.5 MB, maximum 200 rows per request

​
Direct download (download_and_parse_resource)

  • Best for: JSON/JSONL files, large datasets, full data analysis
  • Advantages: Supports more formats (CSV, CSV.GZ, JSON, JSONL), handles larger files
  • Limitations: Downloads the entire file, slower for quick previews
For CSV/XLSX files, start with query_resource_data to preview the data structure with a small page size (20 rows). For large datasets requiring full analysis (>1000 rows), use download_and_parse_resource instead of paginating through many pages.

​
Environment considerations

The get_metrics tool only works in the production environment. The Metrics API does not have a demo/preprod instance.
The server can connect to either:
  • Production: https://www.data.gouv.fr (default)
  • Demo: https://demo.data.gouv.fr (set DATAGOUV_API_ENV=demo)

Build docs developers (and LLMs) love

Get started for freeTalk to us