Dbt MCP

What dbt-labs/dbt-mcp MCP server does

The dbt-labs/dbt-mcp MCP server gives MCP-compatible agents access to dbt project context and operational tools across dbt Core, dbt Fusion, and dbt Platform. It covers two broad use cases: understanding a project and taking supported actions against it. Agents can inspect models, sources, exposures, macros, semantic models, tests, seeds, snapshots, and lineage, then use that context to answer questions or generate work.

For analytics use cases, the server exposes dbt Semantic Layer operations. Agents can list metrics and saved queries, inspect dimensions and entities, retrieve compiled metric SQL, query metrics with filters and grouping, retrieve dimension values, and execute SQL on dbt Platform infrastructure. Natural-language requests can also be converted to SQL using project context.

How it works

The dbt-labs/dbt-mcp MCP server exposes the listed capabilities as MCP tools for clients such as Claude and Cursor. Discovery tools read project and account metadata, while local development tools use project files such as manifest.json for selected lineage and node-detail operations. Other tools connect to dbt Platform services for Semantic Layer queries, administrative operations, Fusion compilation, and platform-based lineage.

The dbt CLI tool group allows an MCP client to invoke operations including parsing, compiling, building, running, testing, showing SQL results, generating docs, cloning nodes, and listing project resources. Administrative tools can list projects and jobs, inspect runs and artifacts, retrieve errors, trigger or retry runs, cancel active runs, and override selected job settings.

Setup and configuration

The repository publishes an experimental dbt-mcp.mcpb bundle with each release. MCPB-aware clients can import the bundle after downloading it from the release assets; installation or inspection uses the Anthropic mcpb CLI documentation. The supplied material does not specify a standalone package-manager command, required environment variables, or credential names.

The README points to dbt documentation for MCP setup and describes examples for connecting the server to Claude or Cursor. It also links to examples for developers building their own agent. The exact authentication and account configuration depend on the dbt capabilities being used and are not defined in the provided material.

Tools and capabilities

Key capability groups include:

• Semantic Layer querying, metric discovery, dimension and entity inspection, and compiled SQL retrieval.
• Project discovery for models, sources, exposures, macros, tests, seeds, snapshots, and semantic models.
• Full lineage, related-model search, model health signals, and model execution history.
• dbt CLI actions for validation, compilation, execution, testing, documentation, and resource listing.
• dbt Platform administration for projects, jobs, runs, artifacts, errors, retries, cancellations, and triggers.
• Code generation for model YAML, source YAML, and staging model SQL.
• Fusion and local column-level lineage, product documentation search, and MCP server version or branch metadata.

Several resource-specific detail and parent/child tools are marked deprecated. The supported replacements are get_node_details for resource details and get_lineage for model relationships.

Limitations and notes

The dbt-labs/dbt-mcp MCP server includes an alpha search tool that is not generally available. The README warns that dbt CLI operations may modify data models, sources, or warehouse objects. Job administration tools can also change run state by triggering, retrying, or canceling runs, so clients should only receive access when those effects are understood.

The server includes both platform-backed and local-development functionality, and the available results depend on the relevant dbt project files or Platform services. Product documentation tools are limited to searching and fetching content from docs.getdbt.com. Deprecated tools remain listed but should be replaced with their newer general-purpose equivalents.