CSV to JSON

What this tool does

This converter reads CSV the way RFC 4180 defines it rather than by splitting on commas. A field wrapped in double quotes may contain the delimiter, a line break, or a doubled quote standing for a literal one, and all three survive the trip intact. The header row becomes the object keys; each following row becomes one object in the array.

The delimiter is detected by test-parsing the start of the file with each candidate and keeping the one that produces a consistent column count — so a file full of commas inside quoted, semicolon-separated fields is read correctly. Ragged rows are padded with null and reported rather than silently dropped.

Common uses

• Turning a spreadsheet export into fixtures or seed data for tests.
• Feeding a CSV report into a script, an API request body or a JSON-only import.
• Inspecting a file whose delimiter or quoting you are not sure about, before automating it.
• Converting a database dump into a shape you can filter with jq.
• Rebuilding structure from flat exports, using the dotted-key option.

A short example

This CSV, with a comma and a doubled quote inside quoted fields:

id,name,active
1042,"Lovelace, Ada",true
1043,"Katherine ""Kat"" Johnson",false

becomes:

[
  { "id": 1042, "name": "Lovelace, Ada", "active": true },
  { "id": 1043, "name": "Katherine \"Kat\" Johnson", "active": false }
]

What type inference will and will not do

CSV has no types — every cell is text — so converting 42 to a number is a guess. It is usually the right guess, and the rules here are deliberately conservative:

• Leading zeros stay text. 01234 is a postcode, an account number or a product code far more often than it is the number 1,234.
• Long integers stay text. Anything over 15 digits cannot round-trip through an IEEE-754 double, so a 19-digit snowflake ID is left exactly as written.
• Only true, false and null convert, case-insensitively. yes, Y, NA and - are conventions, not values, so they stay strings.
• Empty cells become null, not empty strings.

Turn inference off when you need every value to stay exactly as it appeared in the file.

Worth knowing

A leading byte order mark is stripped automatically, because otherwise the first column would be named id and every lookup against id would fail — a bug that costs people a surprising amount of time with Excel exports. Two more things worth remembering: quoted fields containing newlines mean you cannot count records by counting lines, and if your data arrives one record per line already, you may want JSON Lines instead of one big array — it streams, where a single top-level array has to be read whole before anything inside it is valid.