16.1 Database Connections: PostgreSQL, MySQL via MCP

Course: Claude Code - Enterprise Development

Section: Advanced MCP Integrations

Video Length: 3-4 minutes

Presenter: Daniel Treasure

Opening Hook

"Databases are the backbone of enterprise systems, but exposing raw database access to AI is risky. Today, we're using MCP to create secure, read-only database connections that let Claude analyze schema and generate migrations—without ever risking your data."

Key Talking Points

What to say:

"The challenge: developers need Claude to understand schema, generate migrations, analyze query performance—but you can't just hand over database credentials."
"The solution: MCP database adapters create a secure interface with scoped permissions."
"We'll show PostgreSQL first (most common), then MySQL. The pattern is identical."
"Key security principle: read-only roles for analysis, restricted write access for migrations only."

What to show on screen:

Anthropic MCP GitHub repository (database adapters section)
Live psql/mysql client showing database structure
Database user roles configured with minimal privileges
Migration generation workflow in Claude

Demo Plan

[00:00 - 00:30] Setup & Connection 1. Show claude mcp add command for PostgreSQL adapter 2. Display connection string format: postgresql://user:pass@host/db 3. Highlight restricted credentials in .env or secrets manager (don't type live password)

[00:30 - 01:45] Read-Only Queries & Schema Analysis 1. Open Claude Code with PostgreSQL MCP connected 2. Ask Claude: "Analyze the users table schema and suggest indexes" 3. Let Claude execute MCP read operations 4. Show returned schema metadata 5. Live demo: Claude generates SQL analysis based on schema

[01:45 - 03:00] Migration Generation 1. Ask Claude: "Generate a migration to add an updated_at timestamp to users table" 2. Show Claude writing migration file with proper naming 3. Demonstrate: Claude suggests best practices (NOT NULL, DEFAULT CURRENT_TIMESTAMP) 4. Show how migrations are written but NOT auto-executed

[03:00 - 03:30] Security Review 1. Discuss database user roles in .env example 2. Show read-only role vs. migration role separation 3. Emphasize: production databases use different credentials per environment

Code Examples & Commands

# Add PostgreSQL MCP adapter
claude mcp add --transport stdio postgres \
  -- npx -y @anthropic/mcp-postgres \
  --connection-string "postgresql://readonly_user:pass@db.example.com/production"

# Add MySQL MCP adapter
claude mcp add --transport stdio mysql \
  -- npx -y @anthropic/mcp-mysql \
  --connection-string "mysql://readonly_user:pass@db.example.com/production"

Database user setup (PostgreSQL):

-- Create read-only role for analysis
CREATE ROLE claude_readonly WITH LOGIN PASSWORD 'secure_password';
GRANT CONNECT ON DATABASE production TO claude_readonly;
GRANT USAGE ON SCHEMA public TO claude_readonly;
GRANT SELECT ON ALL TABLES IN SCHEMA public TO claude_readonly;

-- Separate migration role (restricted write)
CREATE ROLE claude_migrations WITH LOGIN PASSWORD 'different_password';
GRANT EXECUTE ON FUNCTION run_migrations TO claude_migrations;

Example prompt:

Analyze the current schema for the "orders" table and suggest:
1. Missing indexes for common queries
2. Denormalization opportunities
3. A migration to add an "order_status_updated_at" column

Gotchas & Tips

Gotcha 1: Credentials in Command Line - Never paste production passwords in shell history - Use environment variables or .env files - In demo: use test/staging credentials only

Gotcha 2: Scoped Access Not Enforced by Claude - MCP adapter must enforce permissions at database level - Claude can't "accidentally" write data if database user lacks permission - Always verify database role before connecting

Gotcha 3: Large Schema Analysis - If you have 100+ tables, Claude may struggle - Solution: Narrow MCP scope to specific schemas/tables - Show how to filter in MCP adapter config

Tip 1: Schema Documentation - Claude generates excellent migration comments if given table context - Example: "This table tracks customer orders since 2020. Add a 'metadata' column."

Tip 2: Migration Safety - Always have Claude generate, not execute - Review migrations before running in production - Use MCP for read-only analysis of existing migrations

Tip 3: Performance Queries - Ask Claude: "Show me the most expensive queries on this schema" - Claude analyzes table sizes and indexes to suggest optimizations

Lead-out

"You've now seen how to safely connect Claude to your database layer using MCP. Read-only access for analysis, restricted write for migrations. In the next video, we'll pivot to error monitoring—connecting Claude to Sentry to debug production failures in real time."

Reference URLs

Anthropic MCP PostgreSQL Adapter: https://github.com/anthropics/mcp-servers/tree/main/src/postgres
Anthropic MCP MySQL Adapter: https://github.com/anthropics/mcp-servers/tree/main/src/mysql
PostgreSQL Role Documentation: https://www.postgresql.org/docs/current/database-roles.html
MySQL User Privileges: https://dev.mysql.com/doc/refman/8.0/en/user-account-management.html

Prep Reading

Review the MCP adapter source code for your database (5 min)
Set up test database with sample schema (10 min)
Create read-only user role and test permissions (10 min)

Notes for Daniel

Tone: Emphasize security as the primary motivation—this isn't just about convenience, it's about responsible AI access to critical systems.
Live demo focus: The magic is watching Claude understand schema and generate migrations without touching the actual database. Make that moment clear.
Common question: "Can Claude modify the schema?" Answer: Only if you give the database role that permission—which you shouldn't for analysis-only use cases.
Staging vs. production: If time allows, mention using a staging database for demos (safer for live screen).
Connection timeout: If demo lags, have a pre-recorded fallback showing the output (MCP connections can be slow over network).

Quick Reference