MCP Memory Layer

Example Use Cases

• 1

  Generate accurate SQL queries against complex DuckDB/dbt data warehouses.

• 2

  Leverage curated dbt metadata and lineage to improve query correctness.

• 3

  Learn from past query patterns and schema corrections to avoid repeated mistakes.

Description

This MCP server adds a memory layer between LLMs and DuckDB databases, providing corrections, dbt model context, and query popularity tracking to help LLMs generate more accurate SQL on the first try. It reduces repeated mistakes by capturing schema gotchas, leveraging curated dbt metadata, and learning from query usage patterns. The system supports any DuckDB/dbt project and includes an evaluation framework for A/B testing memory features.