metadata-impact-analysis

name	metadata-impact-analysis
description	Analyze downstream impact before making schema changes. Traces lineage through dbt models and dashboards to identify what breaks if a column is renamed, dropped, or its type changes. Triggers include "impact analysis", "what breaks if", "can I rename", "can I drop", "downstream impact".

Impact Analysis

Analyze the downstream impact of a schema change (rename, drop, type change) on a dbt model column. Traces lineage through OpenMetadata to identify affected models, dashboards, and column references.

How It Works

Execute steps sequentially — complete each step before starting the next. Do NOT fire multiple MCP calls or file reads in parallel.

Step 1: Parse input

Extract model and column from $ARGUMENTS. Accepts:

• campaign_performance.total_conversions → specific column
• campaign_performance → full table (all columns)
• Natural language like "rename total_conversions from campaign_performance" → extract both

Step 2: Confirm entity exists

Call search_metadata to find the table by name. This returns columns inline so you can confirm the column exists and get the correct FQN. Do NOT call get_entity_details at this stage.

Step 3: Trace downstream lineage

Call get_entity_lineage on the confirmed FQN. Collect all downstream nodes (tables, views, dashboards). Present a brief summary to the user before continuing:

"Found N downstream entities: [list]. Checking column references..."

Step 4: Check SQL references

Use Grep to search for the column name across dbt/models/ SQL and YAML files. This is a single local search — no MCP calls needed.

Step 5: Check dashboard impact

If lineage returned dashboard nodes, call search_metadata for dashboards. If no dashboards in lineage, skip this step.

Step 6: Compile report

Combine lineage, SQL grep results, and dashboard data into the output format below. Include ownership from any entity that had it set.

MCP Tools

• get_entity_details — Entity by FQN + entity_type: "table" or "dashboard" (read-only)
• get_entity_lineage — Downstream dependencies from a given entity (read-only)
• search_metadata — Fallback search if FQN lookup fails (read-only)

Reference

FQN pattern: marketing_postgres.postgres.marketing.{model} Mart models: campaign_performance, daily_summary, user_journey, channel_attribution dbt SQL: dbt/models/marts/{model}.sql, dbt/models/intermediate/{model}.sql

Output Format

## Impact Analysis: {model}.{column}

Change type: rename | drop | type change

### Direct downstream
| Entity | Type | Column references | Owner |
|--------|------|-------------------|-------|
| daily_summary | table | total_conversions (3 refs in SQL) | — |

### Dashboard impact
| Dashboard | Service | Granularity |
|-----------|---------|-------------|
| Agentic Data Modeling Demo | Metabase | dashboard-level only |

### Risk: HIGH | MEDIUM | LOW
- HIGH: dashboard or 2+ downstream models affected
- MEDIUM: 1 downstream model affected
- LOW: no downstream references found

### Recommended actions
1. Rename column in {model}.sql
2. Update references in: daily_summary.sql (lines X, Y, Z)
3. Update _marts.yml descriptions
4. Re-run dbt + ingestion
5. Notify: {owner} (owner of downstream assets)