I’m always looking for ways to tighten my workflow. Claude Code has been a productivity rocket booster, but a few clunky, manual steps were still dragging me down. One big one? Giving Claude a full, crystal-clear picture of my database structure.

For quick jobs, I fed it the CREATE TABLE output:

SHOW CREATE TABLE `table_name`;

It worked fine for a handful of tables. But for a big database? It turned into a mind-numbing copy-paste slog.

Next, I tried a complete dump with a query ChatGPT suggested:

SELECT
    table_name,
    column_name,
    column_type,
    is_nullable,
    column_default,
    extra
FROM information_schema.columns
WHERE table_schema = 'matter'
ORDER BY table_name, ordinal_position;

The output looked like this:

table_name	id	int(11)	NO		auto_increment
table_name	uid	varchar(255)	YES	NULL	
table_name	created	timestamp	YES	NULL	
table_name	createdby	varchar(255)	YES	NULL	
table_name	org_id	int(4)	YES	NULL	
table_name	disabled	tinyint(4)	YES	0	

Faster? Yes. Complete? Nope. Missing indexes, charsets, comments—you know, the important bits.

Not good enough.

So I went down the rabbit hole, doing deep-dive research to figure out the best, fastest, AI-friendly way to get the whole picture. Working with both ChatGPT and Claude, I built a lean, best-practice CLI tool to pump out full documentation in seconds.

Grab it here: https://github.com/billnobes/mysql-schema-export — PHP and Python versions included.

The tool takes configs a few different ways and spits out JSON like this:

{
  "database": "mydb",
  "generated_at": "2025-01-15T10:30:00-05:00",
  "schema_version": "2025-01-15",
  "tables": [
    {
      "name": "users",
      "columns": [
        {
          "name": "id",
          "type": "int(11)",
          "nullable": false,
          "default": null,
          "extra": "auto_increment",
          "comment": "User ID",
          "position": 1
        }
      ],
      "primary_key": ["id"],
      "unique": [],
      "indexes": [],
      "foreign_keys": [],
      "table_info": {
        "row_count_est": 1250,
        "engine": "InnoDB",
        "collation": "utf8mb4_general_ci"
      },
      "ddl": "CREATE TABLE `users` (...)"
    }
  ]
}

Yes, the file can get huge—thousands of lines. But after testing, both ChatGPT and Claude agreed: in this case, one monster file beats a pile of tiny ones.

And because not every project needs the whole dump, there’s a table-name filter:

# Export all tables (default)
--filter '.*'

# Export tables starting with 'user_'
--filter 'user_.*'

# Export tables ending with '_log'
--filter '.*_log$'

# Export specific tables
--filter '(users|orders|products)$'

If you’re tired of spoon-feeding your AI assistant table definitions, give it a try.