Metabase MCP Integration

Weldable's Metabase MCP integration gives your AI agents access to your Metabase instance, including saved questions, dashboards, and the ability to run queries against your connected databases. Your agent can pull analytics data, monitor metrics, and generate reports through natural language without anyone opening the Metabase interface.

Metabase is one of the most widely deployed open-source BI tools, with both self-hosted and cloud options. The MCP ecosystem around Metabase has grown rapidly, with community-built servers offering 70+ tools and support for AI-generated SQL, dashboard automation, and natural language querying. Weldable connects to this API surface and handles authentication so your agent can treat Metabase as just another data source in a multi-tool workflow.

Use cases

Natural language data queries

Your agent translates questions like "how many new users signed up last week?" into Metabase native queries or SQL, runs them against your connected database, and returns the results. Metabase's question framework handles the query execution and data formatting. Business users get answers without learning SQL or building custom questions in the Metabase UI.

Dashboard metric monitoring

Your agent polls key dashboards on a schedule and compares current metric values against previous periods or predefined thresholds. When a metric moves significantly (e.g., daily active users drops more than 15% week over week), the agent sends an alert to the relevant team in Slack with the current value, the comparison, and a link to the full dashboard.

Automated reporting

Your agent pulls results from saved questions in Metabase, combines them into a formatted report, and delivers it to stakeholders. A weekly revenue report might pull data from three different saved questions (revenue by product, revenue by region, month-over-month growth), assemble them into a single summary, and post it to the finance channel every Friday at 5pm.

Data exploration for product teams

Product managers can ask your agent questions about user behavior without waiting for an analyst. "What's the conversion rate from signup to first purchase for users who joined through the referral program?" gets translated into the appropriate Metabase query. The agent runs it, returns the result, and can follow up with related cuts of the data in the same conversation.

Database health checks

Your agent runs diagnostic queries through Metabase to check for data pipeline issues: tables that haven't been updated in the expected window, row counts that differ from yesterday by more than a threshold, or null rates in critical columns. It posts a daily data health summary to the engineering channel so issues get caught before they affect downstream reports.

How it works

Connect your Metabase instance by providing the URL and an API key or session credentials through Weldable's setup flow. Metabase authenticates via API keys (available in Metabase Pro and Enterprise) or username/password session tokens. Weldable manages the session lifecycle and handles re-authentication when sessions expire.

Once connected, describe what you need in plain English. Your agent maps requests to the Metabase API, whether that's running a saved question, executing a native query, or pulling dashboard card data. It resolves question and dashboard names to their IDs and returns structured results your agent can summarize or pass to other integrations.

Tips

Saved questions are the most reliable query target. Running a saved question through the API returns consistent, pre-validated results. Ad hoc native queries work too, but saved questions carry the benefit of having been reviewed by your data team. Point your agent at saved questions whenever possible.

Metabase distinguishes between simple and native queries. Simple queries use Metabase's visual query builder (no SQL). Native queries are raw SQL. The API supports both, but native queries require that the authenticated user has native query permissions on the target database.

Collection structure affects discoverability. Metabase organizes questions and dashboards into collections. Keep your collections well-organized with clear names so your agent can locate content by description. A question buried in an unnamed personal collection is harder for the agent to find than one in a shared "Finance Reports" collection.

API keys are the preferred authentication method. If you're running Metabase Pro or Enterprise, create a dedicated API key for Weldable rather than using session-based auth. API keys don't expire on inactivity and don't tie up a user license.

Self-hosted instances need network access. If your Metabase instance runs behind a firewall or VPN, Weldable needs a reachable endpoint. Consider setting up a secure tunnel or allowlisting Weldable's IP range in your network configuration.