Skip to content

Trust & Safety

dbt-tools reads dbt artifacts and turns them into operational signals for humans, scripts, and agents. Before using production artifacts, remote object storage, or AI-agent integrations, review the pages in this section.

Start here

TopicRead this if…
Data boundariesYou want to know what files are read and what data may appear in output
Agent safetyYou are pointing MCP or agent skills at production artifacts
Production hardeningYou are deploying in CI, cloud storage, or shared environments
LicensingYou are evaluating dbt-tools for commercial or internal platform use

Core principles

dbt-tools reads artifacts, not warehouses. The default tools in this repository parse local or remote artifact files. They do not issue queries against your warehouse unless you explicitly configure a feature that does so.

Output reflects artifact content. CLI JSON, Web UI views, and MCP tool responses contain the same data that is in your artifact files. What you expose in artifacts, you expose in tool output.

Agent clients control their own data handling. The MCP server returns data to the connected AI client. That client's own settings govern whether it persists, logs, or transmits the returned data. dbt-tools does not control what an AI client does with tool responses.

Read-only access is sufficient. dbt-tools does not write to artifact files or modify your dbt project. Credential policies should grant read-only access to the artifact prefix.

Released under the repository license terms.