dbt-cloud-log-extractor

name	dbt-cloud-log-extractor
description	Query the dbt Cloud MCP server to extract model run history, pass/fail status, execution times, and failing test details for a given time window. Returns a structured report in chat. Use when asked to check dbt Cloud logs, monitor model runs, review test failures, extract dbt run history, or report on dbt Cloud job health.

dbt Cloud Log Extractor

Extract full model and test execution logs from dbt Cloud.

Full log (default) → run the Python script scripts/fetch_dbt_logs.py — one shot, no MCP spam.
Scoped / ad-hoc query → use the project-0-cursorskills-dbt MCP tools directly.

Prerequisites

• .env contains DBT_ACCOUNT_ID, DBT_HOST
• dbt_project/drip_transformations/mcp.yml has a valid refresh token
  If the script prints "refresh token expired", re-authenticate:

  cd dbt_project/drip_transformations && uvx dbt-mcp auth
  

Project constants

Variable	Value
account_id	70471823552613
prod_env_id	70471823537955

---

Default workflow — full log via Python script

Step 1 — Run the fetcher script

python3 scripts/fetch_dbt_logs.py --runs 10

Options:

• --runs N — number of recent runs to include (default 10)
• --job <id> — limit to a specific job ID

The script outputs JSON to stdout:

{
  "model_runs": [
    {
      "executed_at": "2026-04-28 13:17:19 UTC",
      "job": "Dbt run full refresh",
      "model": "FactSales",
      "status": "success",
      "exec_time": "4.55s",
      "rows_affected": "508",
      "rows_inserted": "508",
      "rows_updated": "—",
      "rows_deleted": "—"
    }
  ],
  "test_runs": [
    {
      "executed_at": "2026-04-28 13:17:25 UTC",
      "job": "Dbt run full refresh",
      "model": "FactSales",
      "test": "not_null_SalesHashPK",
      "status": "pass",
      "exec_time": "1.23s"
    }
  ],
  "meta": { "runs_fetched": 10, "generated_at": "...", "account_id": "..." }
}

Row count fields:

• rows_affected — total rows touched (always populated when available)
• rows_inserted / rows_updated / rows_deleted — only populated for incremental MERGE runs; "—" otherwise

Step 2 — Render in canvas

Read and follow ~/.cursor/skills-cursor/canvas/SKILL.md, then build a canvas with:

1. Stat grid — total model runs, successes, errors, total tests, passes, fails
2. dbt model runs table
3. dbt tests table

Output format

dbt model runs:
| DateTimestamp | Job | Model | Status | Execution Time | Rows Affected | Rows Inserted | Rows Updated | Rows Deleted |

dbt tests:
| DateTimestamp | Job | Model | Test | Status | Execution Time |

• DateTimestamp: YYYY-MM-DD HH:MM:SS UTC
• Model: short name only (FactSales, not model.drip_transformations.FactSales)
• Status: success / error / skipped — apply rowTone (success / danger / warning)
• Execution Time: seconds to 2 dp, e.g. 4.55s
• Rows Affected / Inserted / Updated / Deleted: integer string or — when unavailable

---

Scoped workflow — MCP tools (single model or run)

When the user limits scope (e.g. "show me FactSales" or "last 3 runs"):

get_model_performance(
  unique_id  = "model.drip_transformations.<Name>",
  num_runs   = <N>,
  include_tests = true
)

Skip the Python script entirely. Extract fields directly from MCP response and render canvas.

---

Known models

All belong to drip_transformations. unique_id format: model.drip_transformations.<Name>

Facts: FactSales, FactPurchaseOrder, FactInventorySnapshot
Dims: DimCustomer, DimProduct, DimEmployee, DimStore, DimSupplier, DimWarehouse, DimDate
Views: prefix vw_ on each of the above

Known recurring issues (as of Apr 2026)

• FactPurchaseOrder — BINARY→NUMBER type mismatch on PURCHASEORDERHASHPK; fix with dbt run --full-refresh. Any column data type change requires a full refresh — incremental runs do not alter existing column types and will fail.
• vw_DimEmployee — Department and IsActive nulls from bronze source (data quality)
• Unit tests on vw_DimCustomer / vw_DimEmployee — require schema DBT_UNIT_TEST_RUN in Snowflake