CSV ⇌ JSON converter

Two formats, one table

CSV is what spreadsheets, exports, and data teams hand you: one record per line, fields separated by a delimiter, and a header row if you are lucky. JSON is what APIs, config files, and front-end code expect: named fields, real numbers and booleans, and nesting. Moving between them sounds trivial until a field contains a comma, a quote, or a line break, or a column of zip codes turns into integers and loses its leading zeros.

This converter goes both ways. Paste or drop a file and the direction is picked from the first character: input starting with [ or { is treated as JSON to CSV, anything else as CSV to JSON. Override it when the guess is wrong. Everything runs in your browser, with files above about 1 MB handed to a Web Worker so the tab stays responsive up to the 10 MB limit.

CSV to JSON

id,name,city,zip
1,"Ada, Countess",London,SW1A
2,Linus,Helsinki,00100

[
  { "id": "1", "name": "Ada, Countess", "city": "London", "zip": "SW1A" },
  { "id": "2", "name": "Linus", "city": "Helsinki", "zip": "00100" }
]

With the header option on, the first row becomes the keys of every object. Turn it off and the keys become column_1, column_2, and so on. The array-of-arrays shape skips keys altogether and gives you a plain matrix, which is smaller and convenient when you are feeding a charting library or a spreadsheet API.

Quoted fields with embedded commas, doubled quotes, and newlines inside a cell are handled by PapaParse, the same parser used by most browser-based data tools. Delimiter auto-detection covers commas, semicolons (the default in many European Excel locales), tabs, and pipes; you can also force one from the options.

Type inference, and why it is off by default

CSV has no types: everything is text. JSON does, so a converter has to decide whether 42 is a number and true is a boolean. Guessing is convenient and dangerous in equal measure. A zip code such as 02115 parsed as a number becomes 2115. A 17-digit order id becomes a floating-point approximation, because JSON numbers are IEEE-754 doubles and lose precision past 15 or 16 digits.

That is why inference is opt-in here, and why even with it on the tool refuses to coerce values with a leading zero (other than 0 itself) or more than 15 digits. When you enable it, 42 becomes a number, true and false become booleans, and empty cells become null.

The guard rails are what make a round trip safe. Convert CSV to JSON and back with the same options and the values come out equivalent; only the quoting may differ, because a field is only quoted when it has to be.

sku,qty,active,zip,order_id
A1,3,true,02115,12345678901234567

[
  {
    "sku": "A1",
    "qty": 3,
    "active": true,
    "zip": "02115",
    "order_id": "12345678901234567"
  }
]

What happens to messy files

Real exports are rarely clean, so the converter reports what it fixed instead of failing:

• Duplicate header names get a numeric suffix, so two name columns become name and name_2 rather than one overwriting the other.
• Empty header cells become column_N, using the original column position.
• Short rows are padded with null for the missing fields.
• Rows with more cells than headers keep the overflow in an _extra array instead of silently dropping data.
• A byte-order mark at the start of the file, common in Excel exports, is stripped so the first header does not come out as id.

Each of these shows up as a notice above the output, so you know the shape of the data changed and can decide whether to fix the source.

JSON to CSV

[
  { "id": 1, "user": { "name": "Ada", "email": "ada@x.io" }, "tags": ["admin", "beta"] },
  { "id": 2, "user": { "name": "Linus" }, "tags": [] }
]

id,user.name,user.email,tags
1,Ada,ada@x.io,"[""admin"",""beta""]"
2,Linus,,[]

An array of objects becomes one row per object, with the header built from the union of every key seen, so records with different fields line up under a shared set of columns. Nested objects are flattened to dot-notation columns such as user.name when flatten is on; arrays are kept as a JSON string in a single cell because a flat table has nowhere else to put them. An array of arrays is written as-is, and a single object becomes a one-row table.

The output options match what downstream tools expect: choose the delimiter, switch line endings to CRLF for Windows and older Excel versions, and quote every field if a strict importer needs it. The download adds a byte-order mark so Excel opens accented characters correctly.

Where it fits

• Turning a spreadsheet export into fixture data or a seed file for a test suite, with inference on so numbers are numbers.
• Getting an API response into Excel or Google Sheets for someone who does not read JSON, with flatten on so nested fields become columns.
• Checking what a CSV actually contains before importing it: the row and column summary and the notices catch ragged rows and duplicate headers early.
• Converting semicolon-delimited European exports to plain comma-separated files that a script expects.

From JSON, the JSON, YAML and TypeScript converter takes you on to a typed interface, and the Markdown table generator accepts the same CSV when the destination is a README or a pull request.