Skip to main content
Glama

metabase-mcp

npm version npm downloads CI License: MIT Node Glama score Metabase MCP on mcprush

An MCP server for Metabase with API-key auth and production guardrails. It lets Claude and other AI agents query, explore and build in Metabase on any edition and any version, including from CI and behind a VPN.

Why this one?

Metabase ships an official MCP server. It's good, and if it fits your setup you should use it. This server is for the setups it doesn't fit, mostly open-source and Starter Metabase:

  • Headless on any plan. Authenticates with a Metabase API key, so it works for agents, CI and VPN'd self-hosted instances. The official server's headless auth is Pro/Enterprise only.

  • Governance without an upgrade. Read/write/full modes, per-tool allow/deny lists, column redaction, SQL guardrails, tiered rate limits and a risk-scored audit log, on any edition.

  • Composition. batch_execute runs up to 20 operations in parallel; run_workflow chains steps with output references.

  • AI in the server. NLQ-to-SQL, SQL explain/optimize, insights and trend analysis with your own Anthropic key.

  • Any Metabase version, including pre-v0.60 instances that can't run the official MCP.

Choose the official server if you want OAuth per-user permissions, structured (MBQL) queries or inline charts, or if you're on Pro/Enterprise. → Full comparison

Related MCP server: Metabase MCP Server

Quick start

You need a Metabase URL and an API key (Admin > Settings > API Keys > Create API Key).

Claude Desktop (one-click): download metabase-mcp-*.mcpb from the latest release, double-click it, and enter your URL and key. They're stored in the OS keychain.

Claude Code:

claude mcp add metabase \
  -e METABASE_URL=https://your-metabase.example.com \
  -e METABASE_API_KEY=mb_your_api_key_here \
  -- npx -y @ai-1luvc0d3/metabase-mcp

Any MCP client (for example claude_desktop_config.json):

{
  "mcpServers": {
    "metabase": {
      "command": "npx",
      "args": ["-y", "@ai-1luvc0d3/metabase-mcp"],
      "env": {
        "METABASE_URL": "https://your-metabase.example.com",
        "METABASE_API_KEY": "mb_your_api_key_here",
        "MCP_MODE": "read"
      }
    }
  }
}

Shared server for a team (Docker): run one container and either share one Metabase API key (each person gets a token) or let each person bring their own Metabase credential so Metabase permissions apply per person (MCP_AUTH_MODE=passthrough). Policy, audit log and rate limits live on the server. → docs/docker.md

docker run -d -p 3000:3000 -e METABASE_URL=... -e METABASE_API_KEY=... \
  -e MCP_AUTH_TOKENS=alice:<token>,bob:<token> ghcr.io/1luvc0d3/metabase-mcp:latest
claude mcp add --transport http metabase http://<host>:3000/mcp --header "Authorization: Bearer <token>"

Then ask something like "What dashboards do we have about customer retention?"

Configuration

Variable

Default

Description

METABASE_URL

required

Your Metabase instance URL

METABASE_API_KEY

required

Metabase API key (or METABASE_SESSION_TOKEN)

MCP_MODE

read

read, write or full (see below)

ANTHROPIC_API_KEY

-

Enables the NLQ and insight tools

MCP_TOOLS_ALLOW / MCP_TOOLS_DENY

-

Expose only, or never expose, these tools

MCP_REDACT_COLUMNS

-

Columns to mask in query results, e.g. email,phone*,*_ssn

All options, including timeouts, row limits, rate limits and redaction details: docs/configuration.md.

Tools

Mode

Tools

Adds

read (default)

12

Dashboards, cards, databases, collections, search, SELECT-only SQL, batch_execute, run_workflow

write

22

Create, update and delete cards, dashboards and collections

full

30

AI insights, metric comparison and trend analysis

With ANTHROPIC_API_KEY set, every mode also gets nlq_to_sql, explain_sql, optimize_sql and validate_sql.

Workflows chain tools in one call, passing results between steps:

{
  "steps": [
    { "name": "find", "tool": "search_content", "args": { "query": "revenue", "type": "dashboard" } },
    { "name": "dash", "tool": "get_dashboard", "args": { "dashboard_id": "$find.results[0].id" } },
    { "name": "data", "tool": "execute_card", "args": { "card_id": "$dash.dashcards[0].card_id" } }
  ]
}

Full tool list and more examples: docs/tools.md.

Security

  • SQL guardrails: only SELECT/WITH; DDL, DML and injection patterns are rejected

  • Access control: server modes plus per-tool allow/deny lists, enforced inside batches and workflows too

  • Column redaction: sensitive columns are masked before results reach the AI

  • Rate limits: separate read, write and LLM tiers

  • Audit log: every operation logged with a risk level

  • Secret handling: API keys never reach tool handlers, Metabase errors are sanitized, and auth headers are never forwarded on redirects

Details and limits: docs/security.md. Report vulnerabilities via SECURITY.md.

Privacy policy

  • Collected: nothing. No telemetry, analytics or usage data. Your Metabase URL and keys stay on your machine (in the OS keychain when installed as a Claude Desktop extension).

  • Transmitted: Metabase API calls go straight from your machine to your own Metabase instance. The NLQ and insight tools send your question, schema context and query result samples to the Anthropic API (privacy policy); without an Anthropic key, nothing is sent to Anthropic. Columns in MCP_REDACT_COLUMNS are redacted first.

  • Retained: nothing. Audit logs (if enabled with AUDIT_LOG_FILE) are written only to your local filesystem, with owner-only (0600) permissions.

  • See also the Metabase privacy policy.

Troubleshooting

Connection errors, blocked SQL, rate limits and missing tools: docs/troubleshooting.md.

Contributing

npm install && npm run build   # build
npm test                       # unit + integration tests
npm run lint && npm run type-check

See CONTRIBUTING.md and the Code of Conduct. Bugs and feature requests go in Issues; questions, feedback and production stories in Discussions. If it's useful to you, a ⭐ helps.

License

MIT

Available Tools

12 tools
batch_executeB
Read-only

Execute multiple read operations in a single call. Supports up to 20 operations run in parallel. Supported tools: get_dashboard, get_card, execute_card, get_database_schema, execute_query, search_content, list_dashboards, list_databases, list_cards.

Inputs: operations (required).

execute_cardC
Read-onlyIdempotent

Execute an existing question/card and get results

Inputs: card_id (required), limit, fields, format, +2 more.

execute_queryB
Read-only

Execute a SQL query (SELECT statements only)

Inputs: sql (required), database_id (required), limit, fields, +2 more.

get_cardC
Read-onlyIdempotent

Get details of a specific question/card

Inputs: card_id (required).

get_collectionsA
Read-onlyIdempotent

List all collections

Inputs: no named parameters.

get_dashboardB
Read-onlyIdempotent

Get dashboard details including its cards

Inputs: dashboard_id (required).

get_database_schemaC
Read-onlyIdempotent

Get tables and columns for a database

Inputs: database_id (required), detail, format, tables.

list_cardsA
Read-onlyIdempotent

List questions/cards in Metabase. Returns up to 100 cards. Use search_content for discovery.

Inputs: limit, collection_id.

list_dashboardsA
Read-onlyIdempotent

List all dashboards in Metabase

Inputs: no named parameters.

list_databasesA
Read-onlyIdempotent

List all connected databases

Inputs: no named parameters.

run_workflowA
Read-only

Execute a multi-step workflow pipeline in a single call. Steps run sequentially and can reference previous step results using "$stepName.path" syntax. Example: search for dashboards, then get details of the first result, then execute its cards — all in one call. Max 10 steps.

Inputs: steps (required).

search_contentC
Read-onlyIdempotent

Search across Metabase content (dashboards, cards, collections)

Inputs: query (required), type.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 12 tools

    Disambiguation4/5

    Most tools target distinct resource+action pairs (get_dashboard, get_card, execute_card, execute_query, etc.), and descriptions clarify the split between list_cards and search_content. The main overlap is batch_execute versus run_workflow, which both orchestrate multiple operations, though their parallel-vs-sequential distinction is documented.

    Naming Consistency4/5

    Nearly all tools follow a consistent verb_noun snake_case pattern (get_dashboard, list_cards, execute_query, search_content). The only deviation is batch_execute, which reverses the order to noun_verb, but the overall convention remains predictable.

    Tool Count5/5

    Twelve tools is well-scoped for a Metabase read/query server. Core resources (dashboards, cards, collections, databases, schemas) and execution paths are covered without excessive surface area.

    Completeness3/5

    The server covers discovery, inspection, and execution well, but lacks write/lifecycle operations such as creating or updating cards and dashboards, and has no get_collection detail tool. For a read-oriented analytics MCP this is workable, but notable gaps remain for full content management.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to interact with Metabase analytics platform, allowing them to query databases, manage dashboards and cards, execute SQL queries, and access analytics data through natural language.
      -
    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to interact with Metabase analytics platform, allowing users to query databases, manage dashboards and cards, execute SQL queries, and access analytics data through natural language.
      1
      -
    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to interact with Metabase analytics platform, allowing them to query databases, execute SQL, manage dashboards and cards, and access analytics data through natural language.
      -
    • A
      license
      A
      quality
      B
      maintenance
      A high-performance Model Context Protocol server for AI integration with Metabase analytics platforms. Features response optimization, robust error handling, and comprehensive data access tools.
      6
      630 npm
      89
      MIT