Belgium Data Desk › Use with AI

Ask an AI assistant about Belgian open data

A free, read-only connector that lets Claude or ChatGPT search about 21,700 Belgian open-data records in Dutch, French or English. For common questions it names the official source first (Statbel, the KBO enterprise register, IRCEL, Infrabel...) and the usual pitfalls; for each record it shows the publisher, licence, update cycle and whether the download link worked; and it can read the first rows of a CSV file so the assistant sees the real columns.

Server address

https://data.vecture.be/mcp

No account, no sign-in, nothing to install. The connector only reads catalogue information; it cannot change anything.

Claude

• Claude.ai or Desktop: Settings → Connectors → Add custom connector, then paste the address above.
• Claude Code: run the command below.

claude mcp add --transport http belgium-open-data https://data.vecture.be/mcp

ChatGPT

• Requires a ChatGPT plan that supports custom connectors and developer mode (for example Plus or Pro).
• Settings → Connectors → enable developer mode → add a connector and paste the address above (no authentication).
• In a chat, enable the connector from the tools menu, or use it in deep research.

Other assistants

Any assistant that supports remote MCP servers over HTTP can use the same address.

Things to ask

• “Which datasets show population by municipality, and can I reuse them commercially?”
• “Zoek open data over luchtkwaliteit in Vlaanderen en toon welke links werken.”
• “Quelles données ouvertes existent sur les pistes cyclables à Bruxelles ?”
• “Find the election results by municipality and tell me who to contact about the data.”

What you can trust

• Answers describe catalogue information and link checks. They do not say whether the numbers inside a dataset are correct.
• A record marked ready had a file link that answered with the declared format on the date shown — not a guarantee it works today.
• Topic labels are assigned automatically. Licences come from the publisher; “custom terms” means you must read the publisher's conditions.
• To analyse a file, open the record in Belgium Data Desk: profiling and SQL run inside your own browser.

Tools

• catalogue_overview: counts by kind and topic, and when links were last checked
• search_datasets: multilingual search; the official source comes first for common needs
• get_dataset: publisher, licence, contact, update cycle, columns and link checks
• preview_dataset: column names, types and sample rows from the first 64 KiB of a CSV/TSV file
• search / fetch: the same search and record in the format ChatGPT deep research expects

If something does not work

• “Unknown dataset id”: ids come from search results; ask the assistant to search again first.
• “No previewable CSV/TSV file”: the dataset has no CSV/TSV download in the catalogue, or it is too large or not indexed for preview. Open the landing page or the record in the web app.
• “The publisher answered with a web page instead of a CSV file”: the publisher moved the file. The landing page usually has the new link.
• No results: try fewer words, or the Dutch or French term. Very generic words (data, list) match little.

Support

Questions, wrong records or security issues: info@vecture.be. We answer within two working days. Privacy · Terms