Metabase MCP

Metabase MCP connects an AI agent to a Metabase business intelligence instance through the Model Context Protocol, so an assistant can browse and query analytics content instead of a human clicking through the Metabase UI. The server authenticates against a Metabase deployment, self-hosted or cloud, with an API key or session credentials, and turns the Metabase REST API into MCP tools. From there an agent can list collections, dashboards and saved questions, read the metadata behind them, run a saved card and return its result set, or execute SQL directly against a connected database. The practical value is grounded answers. Because the agent reads the same models, metrics and saved questions your analysts already curated, it can answer a question such as which accounts churned last quarter using the governed definition rather than inventing a join. Server implementations also expose database and table introspection, so the agent can discover schemas before writing a query, and search tools so it can find an existing dashboard before building a new one. It fits analytics and data teams who want a chat client, an IDE assistant or an internal agent doing first-pass analysis, monitoring dashboards, or drafting new questions. Access is bounded by the Metabase credentials you hand it, so permissions, collection scoping and read-only database settings still apply. Some builds are read-only by design, while others also create dashboards, cards and collections.