The dbt MCP Server gives AI agents the context of a dbt project and tools to act on it, across dbt Core, dbt Fusion, and dbt Platform. Agents usually reach it through an agent product such as Claude or Cursor, and example code shows how to connect it to a custom agent. Each release also ships an experimental Model Context Protocol Bundle that MCPB-aware clients import without extra setup.
Its tools fall into groups: SQL execution and text-to-SQL on dbt Platform, the dbt Semantic Layer for metrics and dimensions, the Discovery API for models, sources, lineage, and model health, dbt CLI commands, the Admin API for jobs and runs, code generation for project files, Fusion-based SQL compilation and column-level lineage, and search over the dbt product documentation.
The dbt CLI tools can change data models, sources, and warehouse objects, so they are meant only for clients you trust with that access.
Features
- SQL: executes SQL on dbt Platform with Semantic Layer support and generates SQL from natural language using project context
- Semantic Layer: lists metrics, dimensions, entities, and saved queries, and runs metric queries with filtering and grouping
- Discovery: models, sources with freshness, exposures, bounded lineage graphs, model health and performance, and semantic search for related models
- dbt CLI: build, run, test, compile, parse, list, show, clone, and docs commands, plus lineage from a local manifest
- Admin API: lists, triggers, cancels, and retries job runs, and fetches run errors and artifacts
- Codegen: generates model YAML, source YAML from database schemas, and staging models
- dbt LSP: compiles SQL and traces column-level lineage through the Fusion engine
- Product docs: searches and fetches pages from the dbt documentation site
- MCP Bundle: an experimental bundle per release for MCPB-aware clients
