How to Connect Metabase to ChatGPT

You connect Metabase to ChatGPT by installing Windsor’s native app inside ChatGPT and linking it to your Metabase account through Windsor.ai, no code, CSV exports, or custom connector setup required.

Metabase itself is great for building and viewing dashboards, but it isn’t built for auditing the structure of a whole instance, how many questions and dashboards exist, which collections they live in, who created them, in natural language. Connecting Metabase to ChatGPT through Windsor.ai lets you ask about your Metabase setup directly in chat, and blend that with the other data sources your team already has connected, instead of clicking through screens one at a time.

2 steps to connect Metabase to ChatGPT

Takes under a minute, no code or manual CSV work required.

See the full walkthrough in Windsor’s documentation: How to integrate data into ChatGPT.

Prerequisites

• A Windsor.ai account (free forever plan, no credit card, or paid from $19/month for higher usage)
• A Metabase account or instance with at least one dashboard or saved question already set up
• A ChatGPT account on a plan that supports installing apps

Step 1. Connect Metabase to Windsor.ai

• Create a free Windsor.ai account, or sign in, at onboard.windsor.ai/app/metabase
• Select Metabase from the connector list
• Authorize Windsor to access your Metabase instance
• Connect additional Metabase instances the same way if your team or agency runs more than one

Step 2. Connect the Windsor.ai app in ChatGPT

In ChatGPT, open the connectors or apps settings, search for Windsor.ai, and install it as a native app.

Allow ChatGPT access to your Windsor.ai account when prompted.

Once it’s connected, the Windsor.ai app shows up in ChatGPT ready to query.

Check that the connection worked by asking:

List the data sources connected to my Windsor account. Do you see Metabase there?

If it shows up, you’re ready to query your Metabase instance in natural language.

The metrics that matter for Metabase

Field	What it tells you
dashboards.name	The name of each dashboard in your instance, the starting point for inventorying what’s actually been built.
dashboards.collection_id	Which collection a dashboard lives in, so you can see how dashboards are organized across teams or projects.
dashboards.archived	Whether a dashboard has been archived, useful for spotting outdated dashboards that are no longer maintained.
dashboards.created_at	When a dashboard was first created, letting you track how the instance has grown over time.
cards.name	The name of each saved question, the building block that dashboards are made from.
cards.collection_id	Which collection a saved question belongs to, useful for auditing question sprawl outside their intended folders.
cards.query_type	Whether a question was built with the visual query builder or written as raw SQL, showing how much self-serve versus hand-coded analysis exists.
cards.archived	Whether a saved question has been archived, flagging stale questions still cluttering a collection.
collections.name	The name of each collection, the folders that questions and dashboards are grouped into.
collections.location	A collection’s position in the folder hierarchy, showing how deeply nested your organization scheme is.
databases.name	The name of each database connected to Metabase, the underlying data source that dashboards and questions query against.
users.common_name	Who’s active on the instance, letting you see which team members are creating dashboards, questions, and collections.

What to ask ChatGPT: prompt ideas

Dashboard and question activity

List every dashboard in Metabase along with its collection and creation date, sorted by most recently created.

Show me all Metabase cards flagged as archived, grouped by the collection each one belongs to.

Collection organization

Which collections have the most dashboards and questions, and how deep is the folder hierarchy for each one?

Usage tracking

List the databases connected to Metabase and how many questions query each one.

Show me which users have created the most dashboards and questions in Metabase.

The Metabase connection through Windsor is read-only. ChatGPT can inventory and analyze your dashboards, questions, and collections, but it can’t create, edit, or publish anything back into Metabase.

Other ways to connect Metabase to ChatGPT

Manual export

You can export card and dashboard details from Metabase manually and paste or upload them into ChatGPT. It works, but it’s a static snapshot: every update means exporting again, and it doesn’t refresh as your Metabase instance changes.

Zapier or Make

Zapier or Make can move Metabase data into other tools on a schedule or trigger. That’s a still-maintained workflow, but it requires building and keeping up a separate automation, and it’s a copy of the data rather than a live connection.

Neither approach gives you the same live view inside ChatGPT as the direct Windsor.ai connection.

Conclusion

Connecting Metabase to ChatGPT through Windsor’s native app takes a couple of minutes and gives you a live way to ask about your Metabase instance, its dashboards, cards, collections, and databases, in plain language, alongside any of the other 350+ sources you connect.

🚀 Ready to connect Metabase to ChatGPT? Connect Metabase to Windsor.ai and start asking ChatGPT about your Metabase setup today.