Run cards, dashboards, and SQL queries against a Metabase instance.
4 fields · 2 secret
Credentials are injected into an ephemeral sandbox at boot and destroyed with it.
Standard
Sessions run up to 60 minutes before the sandbox is reclaimed.
60+ AI models
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side.
How models use it and what it is built for.
The Metabase MCP server puts your existing BI layer behind an AI model. Instead of asking the model to guess at your schema, it can list the questions and dashboards your team already built, run them, and read the results.
That distinction matters. A raw database MCP hands the model a schema and hopes it writes correct SQL. Metabase hands it curated, reviewed queries — so "what was revenue last quarter" resolves to the same number your finance team sees, not a fresh join the model invented.
It authenticates with a Metabase username and password against your instance URL, and inherits exactly the permissions of that account.
Concrete operations exposed as tools.
In order. Each step assumes the previous one worked.
The sandbox connects over the public internet. Metabase Cloud works out of the box. A self-hosted instance on a private network needs a tunnel (ngrok, Tailscale Funnel, Cloudflare Tunnel) or an IP allowlist entry.
Do not use your admin login. Create a service user and give it access only to the collections and databases the agent should reach. The MCP server can do exactly what that user can do.
Provide the base URL (https://your-org.metabaseapp.com — no trailing path), plus the service user credentials. All three are injected at sandbox boot.
Ask the model to list available dashboards. If that returns your collections, both the URL and the credentials are right.
What you provide before the sandbox starts.
Metabase API KeyRequiredSecretService-account API key — create at Settings → Authentication → API Keys.
Metabase PasswordRequiredSecretPassword for the Metabase user.
Metabase URLRequiredYour Metabase base URL — e.g. https://your-org.metabaseapp.com.
Metabase UsernameRequiredEmail of the Metabase user the agent will act as.
Paste any of these into Agent Studio once the server is connected.
List every dashboard I have access to and tell me which ones mention revenue.
Run the "Weekly Active Users" question and summarise the trend over the last 8 weeks.
Which tables in the warehouse have a created_at column? Describe their schemas.
Run the signups question, find the biggest week-over-week drop, then query the raw table to explain it.
Known constraints, stated plainly.
Deploy the server hosted, then watch which tools a model actually reaches for — with full JSON input and output on every call. Switch models mid-conversation to compare how each one uses the same server.
About the Metabase MCP server.
Only if the Metabase database connection itself permits writes and the account has native query rights. The safest configuration is a Metabase database connection created with read-only warehouse credentials — then no prompt can cause a write.
Curation. Your saved questions encode business logic — which rows count as active, how revenue is recognised, which test accounts to exclude. A direct database connection loses all of it and the model reinvents definitions on every query.
Yes, and it is the easiest setup — Cloud instances are already publicly reachable, so you only need the URL and a service account.
Any version exposing the standard /api session and card endpoints, which covers all currently supported releases. Very old self-hosted installations may lack some dashboard endpoints.