JSON to CREATE TABLE

infer SQL column types from sample data · DDL for 3 dialects · updated 28 July 2026

Give it a JSON array and it writes the CREATE TABLE statement, choosing each column's type from the values actually present — integers, floats, booleans, ISO timestamps, dates, and JSON for nested structures. Columns that never hold a null get NOT NULL.

How to generate a CREATE TABLE statement from JSON

1. Paste a JSON array. The more records you include, the better the inference — one record makes every column look required.
2. Choose the Dialect. Type names differ: Postgres gets DOUBLE PRECISION and JSONB, MySQL gets DOUBLE and JSON, SQLite gets REAL and TEXT.
3. Leave Add a primary key on. If your data has an id key it becomes the primary key; if not, an auto-incrementing id column is prepended.
4. Check the inferred types in the output pane and adjust anything the sample could not reveal — precision, length limits, foreign keys.
5. Copy or download the DDL, then use JSON → SQL INSERT on the same input to load the rows.

How each type is chosen

• BIGINT — every value in the column is a whole number.
• DOUBLE PRECISION / DOUBLE / REAL — at least one value has a fractional part.
• BOOLEAN (TINYINT(1) on MySQL) — every value is true or false.
• TIMESTAMPTZ / DATETIME — every value matches an ISO 8601 date-time, e.g. 2026-07-28T10:00:00Z.
• DATE — every value matches YYYY-MM-DD.
• JSONB / JSON / TEXT — the value is a nested object or array (or a string that parses as one).
• VARCHAR(n) — MySQL only, when the longest observed string is under 512 characters. The limit is padded by 50% to leave room.
• TEXT — the fallback, including any column whose values mix several of the kinds above.

Where inference cannot help you

Types come from a sample, so treat the output as a first draft. Three things to check by hand:

• Money. A price of 9.5 infers as a float. Financial columns almost always want NUMERIC(12, 2).
• Large integers. IDs beyond 2⁵³ lose precision in JavaScript before this tool ever sees them. If your IDs are big, keep them as strings in JSON and make the column TEXT or NUMERIC.
• NOT NULL. A column is marked NOT NULL only when no sampled record had a null or missing value. A larger sample may disagree.

Privacy

Nothing is uploaded. The conversion runs in this tab using your own browser's JavaScript engine — no server sees your data, and the page keeps working with the network disconnected.

FAQ

Related JSON tools