France MCP Servers 🇫🇷

A curated catalog of Model Context Protocol servers for French data, laws and services. Written entirely in English.

Explore the catalog · Browse the JSON API · Suggest a server · Contribute

About this repository

The Model Context Protocol (MCP) lets AI assistants such as GitHub Copilot, Claude and Cursor connect to external data sources and tools through a standard interface.

This repository is not an MCP server itself. It is an independent, public catalog of MCP projects relevant to France, including open data, legislation, government services and transport. It is not affiliated with or endorsed by the French government or the listed projects.

The infrastructure is adapted from Italia MCP Servers under its original MIT license. This is a separate repository, not a fork; it does not carry the Italian catalog or its Git history.

Start here

Choosing and using a server

1. Choose a category and open a project’s documentation.
2. Check its requirements, license, credentials and available MCP tools.
3. If a remote endpoint is listed, connect with a client supporting the stated transport.
4. Otherwise, follow the project’s installation instructions and add its launch command to your MCP client’s configuration.

Configuration varies between clients and servers. This catalog links to sources; the project’s own documentation remains the authority for setup and usage. A remote link opens an endpoint, not a one-click installation or a health check.

Catalog

The 🎯 Ready score /100 measures documented readiness, not popularity or runtime reliability. Where assessed, click the score for criteria, notes, sources and review date. Unassessed does not mean zero. Entries are sorted by descending score, then by name, with unassessed entries last. Bold names indicate an independent editorial selection. Assessment method and limitations.

Catalog entries have been checked against public repositories and documentation; none has been assigned a Ready score. last_verified records that metadata check, not a runtime test. No third-party MCP server or tool was executed.

📊 Data and Statistics

🗺️ Geospatial and Territory

⚖️ Legal Tech and Law

🏛️ Government and Public Finance

🧭 Public Services

💼 Employment and Labor

🚆 Transport and Mobility

Verification notes

The initial five listings were checked on 2026-09-06 and the six additions on 2026-09-07. Each metadata file links its consulted README at a fixed commit. Licenses reflect upstream declarations; star counts are GitHub snapshots, not recommendations. Installation requirements below come from documentation and source inspection, not execution.

Runtime smoke tests

Community reports are point-in-time observations, not endorsements, certifications or uptime guarantees. On 2026-09-08, xhmq3131 reported that data.gouv.fr MCP worked without authentication using mcporter 0.13.7 on Windows 11 (build 22631) over Streamable HTTP. The endpoint initialized, exposed its tool schemas and returned three results from search_datasets for emploi. A direct MCP protocol check independently confirmed those endpoint behaviors on the same date.

To reproduce the reported smoke test (requires Node.js 24 or newer):

npx --yes [email protected] list https://mcp.data.gouv.fr/mcp --schema
npx --yes [email protected] call https://mcp.data.gouv.fr/mcp.search_datasets \
  query=emploi page_size=3 --output json

Results may change as the catalog data and hosted service evolve. No API key is required for these commands.

The two PISTE servers mention sandbox setup in their READMEs but use production API URLs in source; sandbox compatibility is not verified. Recherche d’Entreprises also documents rate-limit and timeout settings that are not applied to its upstream request. Review these limitations before use. Inclusion does not certify upstream availability, licensing completeness or client compatibility.

Data and API

The machine-readable catalog is regenerated on updates to main by the Pages workflow. Public URLs below become available once the repository maintainer enables GitHub Pages with GitHub Actions and the first deployment succeeds.

curl -s https://bsab.github.io/france-mcp-servers/catalog.json \
  | jq '.servers[] | {name, category, url}'

version identifies the document structure and changes only for incompatible updates. Each servers entry contains the source JSON fields plus url, the canonical project link, and readiness_score, the derived static score or null if unassessed. Optional quality records review evidence and dates; quality_rubric describes how scores are calculated. French catalog category identifiers are English slugs.

Quality and transparency

Inclusion does not certify or endorse a project. Before use, review its code, requested permissions, data handling and license.

To be included, a server must:

• implement the Model Context Protocol, not merely offer an API;
• be relevant to French data, laws or services;
• provide at least one public repository, documentation site or MCP endpoint;
• have sufficient usage documentation.

🎯 Ready score

The Ready score measures documented readiness from 0 to 100. Static rubric v1 evaluates evidence in public documentation:

Each criterion earns zero if not documented in the consulted sources, half its weight if partial, or its full weight if complete. The score is their sum. Unassessed means no assessment is available or sources could not be consulted; it is not zero. Notes and sources, preferably pinned to commits, are public and can be corrected through pull requests. Review dates refer to documentation checks. The starter catalog was researched with AI assistance; its entries are unassessed and open to source-based corrections.

No servers or MCP tools are executed for this review. The score does not certify functionality, security, availability or active maintenance. Well-documented servers may still fail. Weights are an editorial choice, not an experimentally validated benchmark. Assessment is not a prerequisite for inclusion.

Full checklist and assessment contribution guide.

GitHub stars are informational snapshots, not live counts. Stars, featured, remote access and recent activity do not contribute to scores. CI validates the data structure; external links are checked weekly, separately from documentation assessments.

Contributing

The simplest way to help is to suggest a server. To add one directly:

1. Create a kebab-case.json file in servers/.
2. Follow schema/server.schema.json.
3. Validate the entry and regenerate the README.
4. Open a pull request.

python3 -m venv .venv
.venv/bin/python -m pip install -r scripts/requirements.txt
.venv/bin/python scripts/validate_servers.py
.venv/bin/python scripts/build_readme.py
.venv/bin/python scripts/build_readme.py --check
.venv/bin/python -m unittest discover -s scripts -p 'test_*.py'
.venv/bin/python scripts/build_catalog.py

README tables and badges inside BEGIN/END markers are generated; do not edit them by hand. See CONTRIBUTING.md for fields, categories, validation, local preview and deployment instructions.

License and attribution

This catalog is distributed under the MIT license, preserving the original copyright from Italia MCP Servers. The listed servers remain subject to their own licenses.

---

Making French data, laws and services easier to discover for AI assistants.