Skip to content

Metabase MCP Server ★★★ Official

Source: mcpservers.org · Last updated: July 27, 2026 (early morning sweep)
GitHub: metabase/metabase ⭐ 48,400+
Endpoint: https://<your-metabase-instance>/api/mcp (Streamable HTTP, built-in)
Auth: Metabase API key or session token
Category: Business Intelligence / Data & Analytics

Overview

Metabase ships a built-in MCP server starting from its July 2026 release. AI clients connect directly to your Metabase instance and use the semantic layer to search, query, and visualize data — no separate connector required. It builds on Metabase's Agent API to expose tools for navigating your entire BI surface area: databases, tables, questions, dashboards, and collections.

This is the first major BI platform to ship MCP natively — a paradigm shift for how operators interact with business data.

Key Capabilities

  • Search — Find tables, metrics, cards, dashboards, and collections using keywords or natural-language queries
  • Navigate entities — Read metadata for databases, schemas, tables, questions, dashboards, and metrics via metabase:// URIs
  • Build and run queries — Construct queries against tables or metrics, execute them, and get structured results with column metadata
  • Run raw SQL — Execute native SQL against databases (requires native-query permission)
  • Save and update questions — Create or modify saved questions (cards) from agent-constructed queries
  • Dashboard management — Create new dashboards with auto-positioned saved questions, update metadata, archive

Tools Reference

Tool Description
search Search Metabase content by keyword/query
read_resource Read entity metadata using metabase:// URIs
construct_query Build a query against a table or metric
execute_query Execute a constructed query and return results
execute_sql Run native SQL (requires permissions)
create_question Save a query as a question/card
update_question Modify an existing question, including archiving
create_dashboard Build a new dashboard with auto-positioned cards
update_dashboard Modify dashboard metadata or archive

Integration

Prerequisites

  • Metabase instance (self-hosted or Metabase Cloud) running the July 2026+ release
  • API key with appropriate permissions (Admin > Settings > Authentication > API Keys)

1. Claude Desktop

{
  "mcpServers": {
    "metabase": {
      "type": "http",
      "url": "https://metabase.yourcompany.com/api/mcp",
      "headers": {
        "x-api-key": "mb_YOUR_API_KEY"
      }
    }
  }
}

2. Hermes Agent (config.yaml)

mcp:
  servers:
    metabase:
      type: http
      url: https://metabase.yourcompany.com/api/mcp
      headers:
        x-api-key: ${METABASE_API_KEY}

3. Cursor / VS Code

Connect via Streamable HTTP at https://metabase.yourcompany.com/api/mcp with the x-api-key header.

Business Operator Use Cases

  1. Natural Language Analytics — "What was our MRR last month broken down by plan tier?" — agent queries Metabase and returns formatted results
  2. Automated Board Reports — Agent pulls KPIs from Metabase dashboards weekly, formats into a report
  3. Anomaly Detection — Agent monitors key metrics and alerts on unexpected deviations
  4. Ad-Hoc Data Exploration — "Show me churn by acquisition channel for Q2" — agent builds and executes the query
  5. Dashboard Creation — Agent auto-builds dashboards for new initiatives based on natural language descriptions

Pricing

  • Metabase MCP server: Free (included with Metabase)
  • Metabase Open Source: Free (self-hosted)
  • Metabase Cloud: Starter at $85/month (includes MCP support)

Security Considerations

  • API key scoped to specific Metabase permissions
  • Native SQL execution requires explicit permission (disabled by default)
  • All queries respect Metabase's existing data sandboxing and row-level permissions
  • ⚠️ Write operations (create/update questions and dashboards) should be tested in a staging instance first

Verdict

★★★★★ — The first major BI platform to go MCP-native. Essential for any business operator running Metabase who wants AI agents to interact with their BI layer directly. The built-in semantic layer means agents query meaningful business concepts ("MRR", "churn"), not raw table names — this is the right architecture for AI-powered analytics.