Metabase MCP. Audit BI data logic from your agent.

Your AI client's got full control over your Metabase reporting environment. This server lets you audit complex BI setups without ever having to click a single button in the UI. You can talk to your data like an analyst, and your agent handles the heavy lifting behind the scenes.

Dashboard Discovery & Structure:

You'll start by listing every dashboard available using list_dashboards. This gives you a full roster of everything built out on the platform. When you pinpoint the dashboard you want to examine—say, an ID number comes up—your agent pulls the complete layout matrix for that specific report via get_dashboard. This isn't just showing you the finished view; it provides the detailed JSON map.

You see exactly how the board is built: which cards are mapped where, and what filters govern those placements.

Card Logic Audit (The Deep Dive):

To truly understand your metrics, you need to know their source code. Your agent first runs list_cards to pull a raw inventory of every question—every single 'card' defined across the platform. If you find a card that looks right but whose data origin is murky, your client uses get_card. This tool fetches the exact mapping logic or the full SQL query used by that individual BI Card.

You don't guess; you get the source. You can audit hundreds of metrics this way, verifying every single data point.

Global Search and Inventory:

Don't want to search section by section? Your agent runs search_content across everything. It executes keyword searches that hit dashboards, cards, and raw tables simultaneously. If you're looking for 'Q3 sales figures,' it finds every instance of that term anywhere on the platform. For organization boundaries, you can list all structural folders—the Collections—using list_collections.

This shows you the data architecture and how everything is grouped in Metabase.

Data Source Mapping:

Before you audit anything, you need to know what's connected. Your client calls list_databases to give you a definitive inventory of every physical database connection your Metabase instance uses. This tells you exactly where the data lives. You can also use list_cards again to confirm which raw questions are available across all these mapped sources.

Workflow Summary:

When you ask a question, your agent doesn't just guess; it executes a specific sequence of calls. If you need to audit the whole system, you first list all databases with list_databases, then grab a list of every dashboard using list_dashboards. You pick one and get its full layout structure via get_dashboard.

Next, if you want to deep-dive on a metric, you pull the card list with list_cards and run get_card on anything suspicious. For quick checks, your agent uses search_content, letting you find any table or dashboard by keyword across the entire platform. You've got total visibility into what data exists, how it's structured, and exactly which queries are driving every single number.