dbt

dbt Labs' official MCP turns Claude into a surprisingly capable dbt assistant. You point it at your dbt Cloud project via environment variables (host, token, project directory) and it can run models, query the semantic layer in plain English, check which models are stale, and pull documentation directly into your conversation. The semantic layer integration is the standout feature. Instead of writing SQL against your metrics layer, you ask 'what was revenue last quarter by region?' and it translates that into the correct semantic query. It's faster than context-switching to dbt Cloud's UI and genuinely useful during exploratory analysis or stakeholder conversations. I've used it most for drafting tests on new models. You can ask it to suggest tests based on column types and existing patterns in your project, then review the YAML inline. It's not perfect, the suggestions lean generic, but it's a decent starting point that beats staring at a blank schema file. The discovery API access means you can also ask it to surface lineage or find models that depend on a specific source, which is handy during impact analysis. Quirks: it requires dbt Cloud, so if you're running dbt Core locally without the Cloud layer, this won't help. The project directory variable can be fussy if your repo structure doesn't match dbt's expectations. And while it can run models, it's not a replacement for your CI pipeline, it's more for ad-hoc checks or quick validation during development. The documentation search works well but only if your team actually maintains inline docs, which, let's be honest, is hit or miss. Who shouldn't bother: teams not on dbt Cloud, or anyone who doesn't regularly need to interrogate their dbt project from outside the IDE. If you're just running `dbt run` locally and rarely touch the semantic layer, the setup overhead isn't worth it. But if you're already deep in dbt Cloud and want faster access to project metadata or semantic queries without leaving your agent, this is a solid addition.