definite

name

definite

description

Query data, build dashboards, and run automations on Definite via the Definite MCP server. Use whenever the user asks to query their data warehouse, run SQL or Cube queries, create/edit a dashboard or report, set up a data integration or sync, author a semantic model, build a data app, or schedule a Python pipeline. Triggers include "definite", "query my data", "our data warehouse", "dashboard", "run sql", "cube query", "semantic model", "data app", "pipeline", "set up a sync", "connect <source>".

Definite

The Definite MCP server is connected. For any data, analytics, dashboard, or pipeline request, use the Definite MCP tools directly — do not ask the user to "use the connector," just use it.

Querying data

• run_sql_query — row-level detail, ad-hoc joins, schema discovery.
  • DuckLake catalog is uppercase: WHERE table_catalog = 'LAKE'.
  • Three-part table names: LAKE.SCHEMA.table.
  • Always include a LIMIT (100–2000).
  • Discover schema with information_schema, not SHOW TABLES.
  • JSON: -> returns JSON, ->> returns VARCHAR; cast for math: (d ->> 'price')::DOUBLE.
• run_cube_query — aggregated KPIs and time series when a semantic model exists.
  • Field names are camelCase: timeDimensions, not time_dimensions.
  • order must be a list of [name, direction] pairs, not a dict. Right: "order": [["sales.revenue", "desc"]]. Wrong: "order": {"sales.revenue": "desc"} — this fails Pydantic validation.
  • Always set a limit.
  • Worked example:

    {
      "measures": ["sales.revenue"],
      "dimensions": ["sales.region"],
      "timeDimensions": [{"dimension": "sales.order_date", "dateRange": "last 90 days", "granularity": "month"}],
      "order": [["sales.revenue", "desc"]],
      "limit": 1000
    }
    

• search_cube_schema — discover available cubes, dimensions, and measures first.
  • Always call this before run_cube_query. Cube description text can be misleading (e.g. a description that says "use Order Date" when the actual dimension is created_at). Trust the schema, not the prose.

Dashboards, reports, pipelines

• Call get_doc_schema before creating or editing a doc — it returns the full doc.yaml spec. Do not guess the schema.
• create_doc — the whole doc.yaml goes in the raw_yaml string parameter.
• update_doc / update_doc_dataset — edit an existing doc. Use update_doc_dataset to change one dataset without resending the whole YAML.
• list_docs / get_doc / execute_doc — browse, inspect, run.

Integrations & syncs

• collect_secrets (or oauth_connect if it says OAuth is required) → poll with check_credential_session → create_integration. Never ask for secrets in plain chat.
• list_integrations to find sources and the DuckLake destination.
• get_database_schema → configure_sync to set up a scheduled sync.
• list_sync_runs / get_sync_logs to debug a sync.

When stuck

• search_docs searches the Definite product documentation.
• Full agent reference: https://docs.definite.app/mcp/agent-reference

Tool index

Querying: run_sql_query, run_cube_query, search_cube_schema Integrations: collect_secrets, oauth_connect, check_credential_session, create_integration, list_integrations, get_integration, get_database_schema Syncs: configure_sync, list_sync_runs, get_sync_logs Docs: get_doc_schema, list_docs, get_doc, create_doc, update_doc, update_doc_dataset, delete_doc, execute_doc, get_doc_execution_status, get_doc_execution_logs, capture_doc_image Cube models: list_cube_models, get_cube_model, save_cube_model, delete_cube_model Drive: list_drive_files, read_drive_file, write_drive_file, create_drive_folder, delete_drive_file, get_drive_download_url, get_drive_upload_url, list_drive_file_versions, restore_drive_file Docs search: search_docs