metabase-ai-assistant

Metabase AI Assistant is an MCP server that turns natural language prompts into Metabase actions. It provides 134 tools for database querying, dashboard management, user permissions, and metadata analytics, all with structured JSON output to integrate cleanly with downstream workflows. Use it to generate SQL from plain language, auto-create dashboards using templates, explore database schemas, and manage Metabase content programmatically while keeping operations secure with a read-only mode and explicit action approvals. To start, run the MCP server via npx as shown in the quick start, then configure it in your MCP client with the provided metabase server entry. Once running, you can issue natural language requests like “Show total revenue by category for the last 30 days” to trigger ai_sql_generate, or ask the system to create an executive dashboard using mb_dashboard_template_executive. Tools are annotated and categorized (Database Operations, AI-Powered Features, Dashboard Management, User & Permission Management, Metadata Analytics) to help you compose complex workflows with structured JSON outputs for automation.

Prerequisites:

• Node.js (v18+)
• Internet access to fetch packages from npm

1. Install Node.js if not installed. Visit https://nodejs.org/ and download the LTS version for your platform, or use your system package manager.
2. Ensure npm (or corepack) is available:
   • npm --version
3. Run the Quick Start install to start the MCP server:
   • npx metabase-ai-assistant

Optional:

• If you want to pin a specific version, replace npx with npm exec or install the package locally: npm install -g metabase-ai-assistant@latest npx metabase-ai-assistant

Configuration tips:

• In your MCP client, add a server entry named 'metabase' with command 'npx', args ['-y', 'metabase-ai-assistant'], and provide the required environment variables METABASE_URL and METABASE_API_KEY as shown in the example.

Tips and caveats:

• Ensure METABASE_URL points to your Metabase instance and METABASE_API_KEY has appropriate permissions.
• By default, the server can operate in a Read-Only security mode; destructive actions require explicit approval.
• Use the provided MCP annotations in tool outputs to understand each tool's purpose and data flow.
• If you encounter network or authentication issues, check environment variable values and ensure Metabase is reachable from the environment running the MCP server.
• For production deployments, consider enabling auto-backup prompts and configuring activity logging and metadata analytics as part of your observability stack.