Generic fake-data tools hand you isolated rows and leave you hand-wiring the relationships. This generates referenced tables first and points the FK columns at keys that exist, so the INSERT statements run as written. Running analyze_schema first tells you whether it understood your DDL before you generate anything from it.
A test-data generator that reads your real schema. Give it Postgres or MySQL CREATE TABLE DDL, or a JSON schema, and it generates rows in topological order so referenced tables come first and foreign key columns point at primary keys that were actually generated.
- A schema parsed and reported back as tables, columns, relationships and generation order, so you can confirm it read yours correctly before generating anything (analyze_schema)
- Coherent rows generated, as SQL INSERT statements by default, or JSON or CSV (generate_seed_data)
- Foreign keys resolved against generated primary keys, unique columns kept unique, primary keys unique, and NOT NULL honoured
- Realistic values for names, emails, addresses, companies, dates, decimals and enums
- Per-table row counts, enums and value ranges controlled through the JSON schema form
Npx -y seedweaver-mcp and nothing else. It works from DDL or a JSON schema you paste in, not from a live connection, so it never touches your database.
One command — npx -y seedweaver-mcp