seedgen

seedgen

Test fixtures and database seed rows have a habit of being either hand-typed (small, stale, and quietly unrepresentative) or randomly generated (different on every run, so a failing test can never be reproduced). seedgen takes the second path without the drawback: you declare the shape of a record — field names and types, bounds, lengths, choices, date ranges — plus one numeric seed, and it produces exactly the records that seed implies. Run it again next week, on another machine, in CI: the output is identical, because the generator is a pinned PRNG (mulberry32) whose draws are consumed in a pinned order — record by record, field by field in the order you declared them.

Seven field types cover the usual fixture needs:

• integer — whole numbers between min and max, both bounds included (defaults 0 and 100)
• number — decimals between min and max (defaults 0 and 1), rounded to 4 decimal places
• boolean — true or false
• string — lowercase letters, length characters (default 8, at most 64)
• enum — one of the values you supply
• email — an 8-letter local part at @example.com
• date — a real calendar date from from to to inclusive (defaults cover 2020), computed in UTC and formatted YYYY-MM-DD

Properties a field's type does not use are ignored rather than rejected — a stray min on a boolean field changes nothing. Genuinely malformed input is a validation error naming the offending path (for example fields[1].values must be a non-empty array): a seed outside the unsigned 32-bit range, a count outside 1..1000, a duplicate field name, an integer bound that is not an integer, an inverted range, a made-up calendar date such as 2021-02-29. On any error, nothing is generated — there are no partial record sets. The computation is pure and local: no network, no configuration, no files read beyond the input you pass in.

CLI

npx @reneturcios/seedgen generate --input args.json
npx @reneturcios/seedgen --version

args.json contains the full arguments object:

{
  "seed": 42,
  "count": 3,
  "fields": [
    { "name": "id", "type": "integer", "min": 1, "max": 10 },
    { "name": "code", "type": "string", "length": 4 },
    { "name": "active", "type": "boolean" }
  ]
}

The generated set is printed as JSON — {seed, count, records}, each record holding the declared fields in declaration order:

{
  "seed": 42,
  "count": 3,
  "records": [
    { "id": 7, "code": "lwre", "active": false },
    { "id": 3, "code": "qwmg", "active": false },
    { "id": 8, "code": "hfnr", "active": false }
  ]
}

Exit codes: 0 on success, 1 on a runtime error (including an unreadable or invalid input file, or invalid input), 2 on a usage error.

MCP server

npx -y @reneturcios/seedgen mcp

MCP client config:

{
  "mcpServers": {
    "seedgen": {
      "command": "npx",
      "args": ["-y", "@reneturcios/seedgen", "mcp"]
    }
  }
}

Tools

• generate_seed_data — Generate deterministic seed records from field specs and a numeric seed.

  • seed (integer, required) — unsigned 32-bit seed for the pinned PRNG
  • count (integer, required) — number of records to generate (1..1000)
  • fields (array, required, at least one) — field specs as {name, type, ...} objects; name is unique across the specs, type is one of integer, number, boolean, string, enum, email, date, and the remaining properties (min, max, length, values, from, to) tune the type that uses them

  Returns {seed, count, records}.

Docker

docker run -i ghcr.io/cloudcorprecords/seedgen
docker run --rm -v "$PWD:/w" ghcr.io/cloudcorprecords/seedgen generate --input /w/args.json

Development

npm install
npm run build   # tsc -> dist/
npm test        # tsx --test test/*.test.ts (pure local, no network)
npm run dev     # run the CLI from source

License

MIT — Copyright (c) 2026 Rene Turcios