mcp-database

MCP Database Workflow

Use this workflow when you need to query, investigate, or modify database state.

Prerequisites

Gateway instructions already tell you WHICH tools handle database operations. This skill teaches the safe workflow for using them.

Workflow

Step 1: Understand the schema

Before writing any query:

1. List available tables to understand the data model
2. Describe the specific table(s) you'll be working with
3. Note column types, constraints, and relationships

Never write a query against a table you haven't inspected.

Step 2: Read before write

Always run a SELECT to understand current state before any mutation:

• Check what data exists
• Verify your WHERE clause matches the expected rows
• Confirm row counts

Step 3: Execute with appropriate caution

Operation	Caution Level
SELECT	Auto-execute OK
INSERT (new data)	Confirm with user — describe what will be inserted
UPDATE	Confirm with user — show the SELECT of affected rows first
DELETE	Always confirm — show affected rows and ask explicitly
DDL (ALTER, DROP)	Always confirm — these are irreversible

Step 4: Verify after mutation

After any INSERT/UPDATE/DELETE:

1. Run a SELECT to confirm the change took effect
2. Check row counts match expectations
3. Report the before/after state to the user

Decision Points

Situation	Action
Need to understand data model	List tables → describe relevant tables
Simple data lookup	SELECT directly (schema check optional for known tables)
Data modification needed	SELECT first → confirm with user → execute → verify
Debugging a data issue	Schema check → SELECT to inspect state → trace the problem
Need to check logs	Use the Gateway's log retrieval tools

Anti-patterns

• Writing queries without checking the schema first
• Running UPDATE/DELETE without previewing affected rows
• Not verifying mutations after execution
• Assuming table structure from memory (schemas change)