Skip to main content
The dbt Discovery Tools let DinoAI answer questions about your dbt project’s structure, health, and performance from metadata, without reading files or querying the warehouse. They are backed by the artifacts of your Bolt runs, the same data behind the Discovery API. Tool names, parameters, and output mirror the dbt-mcp Discovery toolset, so prompts written against dbt-mcp work unchanged.

​
Capabilities

All nine tools are read-only and run without prompting for approval. A unique_id with the wrong project prefix, for example one guessed from the repository name, still resolves when exactly one node of that type and name exists in another project; the result carries a note naming the node it used (on get_model_performance, on the newest run). When several projects have that name, or none does, the error lists the candidates to retry with.
Discovery data comes from your Bolt runs in the workspace’s default Bolt environment. If a workspace has no Bolt runs with dbt™ artifacts yet, the tools return an error explaining that no discovery data has been ingested.

​
Example Use Cases

Checking model health before a change Prompt
Result: DinoAI reports the model’s last run status, each attached test’s status, and whether its direct parents ran successfully and have fresh sources. Investigating slow runs Prompt
Result: DinoAI returns the model’s historical execution times so you can spot regressions.

​
Working with Other Tools

  • Combine with the Catalog Tool to first find an asset by business concept, then pull its full details and lineage with get_node_details and get_lineage
  • Combine with the SQL Execution Tool to validate the data behind a model whose metadata looks suspicious
  • Combine with the Bolt Logs Tool to jump from a failing model’s health report to the logs of the run that broke it