Analyze Amazon Redshift and read query plans with Claude Code

Watch Claude Code analyze Amazon Redshift data and fix a distribution key problem from the query plan.

Who it's for: Data engineers and analysts who want Claude Code working in their own Amazon Redshift warehouse.

The problem: Analysis SQL written one query at a time can take an afternoon. On a small warehouse, a badly designed query still runs in milliseconds, hiding a problem that grows with the data.

Claude Code connects through the Redshift Data API CLI and the read-only AWS Labs Redshift MCP server. It answers a revenue question in under five minutes and avoids a join that would count revenue twice. From the EXPLAIN plan, it redistributes order_items on order_id and cuts the estimated cost about 16 times.

Takeaways:

  • Use the MCP server for reads, the CLI for writes.
  • Give the agent context on your tables and metrics.
  • Have it read the plan for your slowest query.
  • Read the SQL before you trust the answer.

Resources