AI 開發工具
dbt Charts Rewrites Dashboards as YAML, Enabling AI Agents to Generate, Validate, and Version-Control Them
dbt Labs has released dbt Charts, replacing drag-and-drop BI editing with YAML, SQL, and Git while offering a CLI, workflow Skills, and an optional MCP server. The initial release still cannot query the dbt Semantic Layer directly, and sharing an environment with dbt v2 may cause package downgrades.
dbt Labs released dbt Charts on September 14, placing queries, charts, filters, and layouts in a single YAML file that an engine compiles into an interactive dashboard. Unlike traditional BI tools, which store state in databases or proprietary formats, these files can be committed to Git alongside dbt models, reviewed through diffs and code review, and validated in CI. This also makes them better suited to iterative generation and revision by coding agents. The [official documentation](https://docs.dbtcharts.com/) lists components including bar charts, line charts, maps, facets, and layers. Data can come from inline values, CSV or Parquet files, or SQL queries run against sources such as DuckDB, Postgres, Snowflake, BigQuery, and Databricks.
The `dct` CLI can initialize, validate, render, search, and perform impact analysis. After optional dependencies are installed, `dct mcp serve` can also expose these capabilities to external agents. When initializing an existing dbt project, the tool copies workflow Skills designed for use with Codex, Copilot, Cursor, or Claude Code. The technical value goes beyond simply having “AI create charts”: agent output is constrained to a declarative interface that has a schema, is reproducible, and can be checked before changes are merged.
Several important limitations remain. Dashboards can query only the tables built by dbt through SQL; using the dbt Semantic Layer directly by metric name remains an [unscheduled issue](https://github.com/dbt-labs/dbt-charts/issues/1). As a result, the same business metric may be defined repeatedly across multiple boards and gradually drift. The dbt Charts warehouse adapters also depend on dbt Core 1.x. Installing them in the same Python environment as dbt v2 may silently replace v2, so the official recommendation is to isolate the installation with `uv tool install` or `pipx`. Teams should next evaluate the stability of the YAML schema, the cost of running queries in CI, the scope of MCP permissions, and whether Semantic Layer integration can truly eliminate duplicated SQL.