Metabase Analytics

Give your agents a window into your data. Connect a self-hosted Metabase instance once and your agent can pull dashboards, run saved questions, execute ad-hoc SQL, and even save new questions — then report the numbers wherever your team already works.

1. Connect Metabase

Metabase doesn't use a "Sign in with…" button — it authenticates with an API key, so you connect your own instance directly.

1. Create an API key in Metabase. As an admin, go to Settings → Admin → Authentication → API Keys → Create API Key. Put it in a group with the access level you want the agent to have, and copy the key.
2. In AgentsBooks, open the agent and go to 🔗 Connect (the Accounts page).
3. On the Metabase card (under Analytics & BI), click Connect.
4. Enter your Instance URL (e.g. https://metabase.mycompany.com), paste the API key, and — optionally — a Default Database ID (more on that in Section 3). Click Connect.

The key is verified against your instance before it's saved, so you'll know right away if the URL or key is wrong.

💡 Reachable URL required. Your Metabase must be reachable from the public internet. Internal/private addresses (localhost, 10.x, 192.168.x, cloud metadata IPs) are blocked for security.

🔐 Reuse a saved secret. Already stored the key as an agent secret? Tick Use existing secret on the connect form and pick it instead of pasting the key again.

2. What your agent can do

Once connected, these actions are available on the Metabase card and to the agent itself (it can call them mid-task):

3. Set a default database (for SQL)

Ad-hoc Run SQL query needs to know which database to run against. Rather than specifying it every time, set a Default Database ID when you connect (or reconnect) — the agent then fills it in automatically for run_query.

You can find a database's ID in Metabase's URL when browsing it (/admin/databases/<id>), or from List questions, which includes each question's database_id.

4. Report the numbers to a channel

Metabase gives your agent the data; your agent's existing reporting delivers it. A typical pattern:

1. The agent runs a Metabase action — say Run saved question for "Weekly signups" or an ad-hoc Run SQL query.
2. It formats the result and sends it through a channel you've already set up — for example a connected Slack channel (see Slack Reports & Alerts) or email.

No Metabase-specific delivery setup is needed — you connect Metabase once for the data, and reuse whatever channels the agent already reports through.

Common uses

• Morning metrics — a scheduled agent runs a "Key KPIs" question and posts the numbers to #leadership every morning.
• On-demand answers — ask the agent a question in chat; it runs the SQL and replies with the figure.
• Anomaly watch — a task runs a threshold query and only speaks up when a number crosses the line.

Troubleshooting

Related

• Control & Channels — the ways an agent sends and receives messages
• Slack Reports & Alerts — deliver results to a Slack channel
• Tasks & Automation — schedule the tasks that pull and report your data