Ask for a number and query_metrics answers it from the Semantic Layer; ask why it moved and get_lineage, get_model_health and get_model_performance walk back to the model and the run behind it, with trigger_job_run and retry_job_run available once you know what to rebuild. The tool list is dbt platform APIs and documentation rather than dbt commands run against a local project. A block of the older detail tools — get_model_details, get_exposure_details, get_source_details, get_macro_details, get_seed_details, get_semantic_model_details, get_snapshot_details and get_test_details — carries a DEPRECATED marker pointing at get_node_details, and get_model_parents and get_model_children point at get_lineage; all of them say they will be removed in a future release.
A 38-tool server over a dbt platform account — the Semantic Layer for metric queries, the Discovery API for models and lineage, the Admin API for jobs and runs, and search over the dbt documentation at docs.getdbt.com.
- Metrics from the Semantic Layer listed and searched, then queried by dimension and entity with grouping, filtering and ordering — and the same query returned as compiled SQL without being executed
- Dimension values enumerated before you write a where clause, and saved queries discovered rather than rebuilt
- Every model and mart in the environment, full node details for a model, source, exposure, test, seed, snapshot, macro or semantic model, and the lineage graph in either direction to a chosen depth
- A model's health — last run, last test execution status, upstream freshness — and its execution time and status across historical runs, with test history when include_tests is set
- Dbt jobs and runs from the Admin API: trigger with a git branch, SHA, schema or steps override, cancel a queued or running one, retry a failed one from the point of failure
- A finished run opened up: its artifacts, including manifest.json, catalog.json and run_results.json, and a focused error and warning report instead of the full run log
- The dbt product documentation searched by query and up to 5 pages fetched per call, with an optional query that returns only the relevant sections
Access to a dbt platform account — the tools address its projects, environments, jobs and runs through the Admin and Discovery APIs — and a configured dbt Semantic Layer for the metric tools to read.
One command — uvx dbt-mcp