Metabase to MCP: Saved Questions, Dashboards, SQL and MBQL Queries and Collections in any AI Agent

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Metabase, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Metabase

Work with Metabase, self-hosted or Metabase Cloud, from any AI agent: search and run saved questions and dashboard cards, run SQL or MBQL queries, read database, table and field metadata, browse collections, and create or update questions, dashboards and collections. API key with the permissions of one Metabase group.

Setup

Setting up takes about 3 minutes. You need a Metabase instance (self-hosted or Metabase Cloud) and an admin who can create an API key.

1. Put the address you open Metabase at into METABASE_URL, for example https://metabase.example.com or https://yourcompany.metabaseapp.com, without /api and without a trailing slash.
2. As an admin open Admin settings, Settings, Authentication, API keys, click Create API key, give it a name and a group. The key acts with that group's permissions: view access to the collections and query access to the databases you want is enough for reading; creating and changing content needs curate access. Admin-only reads such as metabase_list_users need the Administrators group.
3. Copy the key (it is shown once) into METABASE_API_KEY. It is sent in the x-api-key header.
4. Install the connector and run metabase_get_current_user: it shows the account the key acts as and confirms address and key.

What it reads: content through metabase_search (questions, models, metrics, dashboards, collections, tables, databases, segments, optionally the SQL text of native questions or the trash), collections and their items, saved questions, models and metrics with the query behind them, dashboards with tabs, filters and cards, and the recent and most popular items. Databases connected to Metabase with schemas, tables and fields (connection details are left out), and the cached values of category fields for filters. People and permission groups.

What it runs: metabase_run_card runs a saved question or model, with parameters for its filters; metabase_run_dashboard_card runs one card of a dashboard with the dashboard's filter values. metabase_run_native_query runs SQL in the database's own dialect, with {{name}} variables through template_tags and parameters; metabase_run_query runs an MBQL query built from table and field ids. Nothing is saved. The AI answers with saved questions first and writes ad hoc queries when nothing fits.

What it writes: metabase_create_card saves a new question, model or metric in a collection or inside a dashboard, metabase_update_card changes one, metabase_create_dashboard creates an empty dashboard and metabase_update_dashboard changes name, filters, tabs or cards. Its dashcards list replaces all cards, so the AI reads the dashboard first, keeps the existing cards with their ids and adds new ones with a negative id on the 24-column grid. metabase_create_collection and metabase_update_collection create, rename and move collections.

Switched off at install: nothing; every tool is on. There are no delete tools: archiving a question, dashboard or collection through the update tools moves it to the trash, where it can be restored with archived false.

Data formats: query results come as data.cols (name, display_name, base_type such as type/Integer or type/DateTime) and data.rows (arrays in column order), with row_count. Dates are ISO 8601 in the instance's report time zone. Ids of databases, tables, fields, questions and dashboards are integers; the top collection is root. A card's type is question, model or metric; in search and collection items a model shows as dataset.

Paging: metabase_search, metabase_list_collection_items and metabase_list_users take limit and offset and return total. The other lists return everything at once. metabase_get_database_metadata can be large for big databases; metabase_list_schema_tables and metabase_get_table_metadata are lighter.

Limits: query tools return at most 2,000 rows, the API's limit for ad hoc results, so the AI aggregates or filters in the query instead of paging through raw rows. A native query runs with the database connection Metabase uses: for databases that must never change, give Metabase a read-only database user.

Common errors: a query that fails comes back with status: failed and the database's message in error, reported as a failed call. 401 means a wrong or deleted key. 403 means the key's group lacks the permission (data access, curate or admin). 404 is a wrong id, or an address without /api routes: check METABASE_URL.

The connector does not delete content permanently, does not change users, groups or permissions, and does not add or edit database connections.

Prerequisites: METABASE_URL, METABASE_API_KEY

Install the connector locally

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000/connectors/store, pick Metabase and paste in the env vars listed under Prerequisites.

Available tools

Example prompts

• "Find the saved question about monthly revenue and run it for the last 12 months."
• "Which dashboards are viewed most in our Metabase? List them with their collection."
• "Run the cards of the Sales overview dashboard with the region filter set to EMEA and summarise the numbers."
• "Write a SQL query on the orders table that counts orders per country this year, and show me the result."
• "Save the query you just ran as a bar chart question in the Marketing collection."

Next steps

• Connect to Claude
• Connect to ChatGPT