dbt MCP Server: Give AI Agents Access to Models, Tests, and Lineage

Why Give AI Access to dbt?

dbt transforms raw data into analytics-ready models. When models fail, tests break, or you need to understand upstream dependencies, you switch to dbt Cloud to check run status, browse lineage graphs, and read test results. Each switch breaks your coding flow.

An MCP server for dbt lets your AI agent check model status, view lineage, inspect test failures, and query run history directly from Claude, Cursor, or Windsurf through DataFaucet.

What Becomes Possible

Once your AI agent has dbt access:

• "Which models failed in the last production run?"
• "Show me the lineage for the orders_fact model"
• "What tests are failing on the customer_dim table?"
• "When was the last successful full refresh of the revenue model?"
• "What models depend on the raw_events source?"
• "Show me the SQL for the latest compiled version of order_metrics"

How It Works

1. Go to DataFaucet and enter your dbt Cloud URL (cloud.getdbt.com)
2. Authenticate with your dbt Cloud credentials
3. Browse models, check runs, view lineage as you normally would
4. Each API call becomes an MCP tool your AI can invoke
5. Deploy and connect to Claude, Cursor, or Windsurf

The AI agent accesses whatever projects and environments your dbt permissions allow.

Security Best Practices

• Read-only API tokens — use dbt Cloud service tokens with read-only scopes
• No run triggers — prevent AI from starting dbt runs or deployments
• Project-scoped access — limit to specific dbt projects, not entire account
• No environment variables — exclude access to connection credentials
• No source freshness mutations — block any write operations
• Audit via dbt Cloud logs — all API access is tracked

Use Cases by Role

Analytics engineers: Model debugging. "Why did the orders model fail? Show me the test results and upstream dependencies" while writing the fix in your editor.

Data engineers: Pipeline coordination. "Which dbt models ran after the Airflow extract finished? Are all downstream models fresh?"

Data analysts: Self-serve lineage. "What tables power the revenue dashboard? When was the data last refreshed? Any tests failing upstream?"

Engineering managers: Data quality overview. "How many model failures this week? Which models are most unreliable? What's the average run time trending?"

dbt Cloud UI vs AI Agent

Integration with Data Stack

dbt connects to Snowflake, BigQuery, Airflow, and Looker. With DataFaucet MCP servers for each:

• Check if upstream Airflow DAGs succeeded before investigating dbt failures
• Correlate dbt model freshness with Snowflake query performance
• Verify Looker dashboards reference up-to-date dbt models

Getting Started

Set up read access for your AI agent:

1. Create a dbt Cloud service token with read-only permissions:

- Jobs: Read (run history and status)

- Projects: Read (model definitions and lineage)

- Environments: Read (deployment status)

- No run or admin permissions

1. Point DataFaucet at your dbt Cloud instance
2. Authenticate with the service token
3. Browse models and runs as you normally would

Related: Connect dbt to AI for the quick-start guide, Airflow MCP Server for pipeline orchestration, Snowflake MCP Server for warehouse access.