// the find
dbt-labs/dbt-mcp
A MCP (Model Context Protocol) server for interacting with dbt.
An MCP server from dbt Labs that exposes dbt Core, dbt Fusion, and dbt Platform capabilities as tools for AI agents — things like querying the Semantic Layer, inspecting model lineage, triggering job runs, and generating boilerplate YAML. It's aimed at data engineers who want to wire Claude or Cursor into their dbt workflow without writing glue code themselves.
The tool surface is genuinely wide and well-organized: Semantic Layer queries, Discovery API, Admin API, CLI passthrough, LSP-backed column-level lineage, and even product doc search are all separate namespaces rather than one giant blob. The MCPB bundle format means supported clients can install it without touching Python or environment config. It ships with working examples for six different agent frameworks (LangGraph, CrewAI, OpenAI, ag2, Pydantic AI, Google ADK), which is more than most MCP servers bother with. Changie-based changelog discipline and OpenSSF badge suggest they're treating this as a real product, not a weekend hack.
The CLI tools (run, build, test) are a footgun by design — the README explicitly warns they can modify warehouse objects, but there's no sandboxing or dry-run enforcement in the tool layer itself; you're trusting the agent not to call dbt run on prod. The dbt Platform features (Semantic Layer, Discovery API, Admin API, Fusion) require a paid dbt Cloud account, so the open-source half of the tool list is thinner than the README implies. Column-level lineage locally requires the dbt Labs VS Code extension's LSP binary, which is an undocumented dependency that will silently fail if you don't know to look for it. Version is at 1.20.x but the project only started in late 2024 — that cadence suggests instability in the interface, and pinned-only dependency updates mean you may be sitting on old transitive vulnerabilities between security patches.