- VALID
- ERROR
- FILTER
- HUMAN
Short answer
How can I check if my Metabase metrics are accurate?
Your agent pulls the underlying SQL and filter configurations directly from specific Metabase cards. It then compares that code against your expected business logic to find discrepancies. You get a clear breakdown of where the math or the filters might be wrong.
Audit outcomes.
Where your findings land.
The agent processes your request and categorizes the results.
VALID
Confirmed logic
The agent confirms the SQL matches your requirements. You get a summary of the verified filters and calculations.
ERROR
Logic mismatches
The agent identifies where the SQL code deviates from your intent. It points out specific lines in the query that cause issues.
FILTER
Filter discrepancies
The agent flags hidden or incorrect dashboard filters. It shows how these filters are modifying your raw data results.
HUMAN
Manual review needed
The agent flags ambiguous queries that it cannot confidently interpret. You'll need to step in and check these manually.
The workflow
What your AI does when the request arrives.
The agent moves from finding the data to analyzing the logic.
Locate the card
The agent finds the specific dashboard element you want to check.
list_cardsExtract the code
The agent pulls the exact SQL query and filter settings used by that card.
get_cardCompare logic
The agent compares the extracted SQL against the business rules you provided in your prompt.
get_cardReport findings
The agent presents the discrepancies or confirms the code is correct.
list_cardsget_card
Try it
Copy these to start.
Use these prompts to begin your audit.
Starting points
Replace the card names and specific metric descriptions with your own details.
Metabase Connector
You're all set. Choose your MCP client and follow the setup instructions.
https://edge.vinkius.com/vk_preview_81usxhuffmH5qVmHtpmbFZWoVXrKKHjT3u2k9g8R/mcpClaude Desktop
Follow the steps below to connect in seconds.
- 1In Claude Desktop, open Settings → Connectors.
- 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
- 3Click Add and start a new chat — Metabase capabilities are ready to use.
{
"mcpServers": {
"metabase-business-intelligence-analytics-mcp": {
"url": "https://edge.vinkius.com/vk_preview_81usxhuffmH5qVmHtpmbFZWoVXrKKHjT3u2k9g8R/mcp"
}
}
}Check the SQL for the 'Monthly Active Users' card to see if it's filtering out guest accounts.
List all cards in my 'Sales Dashboard' and tell me which ones use a hardcoded date filter.
Look at the 'Revenue by Region' card. Does the SQL correctly group by the 'country_code' column?
Verify the logic in the 'Churn Rate' card. I need to make sure it only counts users from the last 30 days.
Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline
Start here
Connect Metabase once, then ask.
Just link your Metabase instance. Your credentials stay encrypted, and you can immediately start asking questions about your cards.
Connect Metabase to your AIFAQ
How this task behaves.
- 01
Can the AI change the SQL in my Metabase cards?
No. The agent can only read the code and filters. You must make any changes manually in Metabase.
- 02
Can the AI delete any of my dashboards?
No. This MCP only has read access to your cards and their configurations.
- 03
How does the AI know if the SQL is 'wrong'?
The AI relies on the business logic you provide in your prompt. It compares what the SQL actually does against what you say it should do.
- 04
Does it work with GUI-based questions?
Yes, if the question can be answered by inspecting the card's metadata and SQL, the agent can handle it.
- 05
Can I audit an entire dashboard at once?
Yes. You can ask the agent to list the cards in a specific dashboard and then audit them one by one.
More questions about Metabase? See everything the Metabase Connector can do
Connect Metabase to Claude, Cursor, ChatGPT & more
