Question 1

Why does my JSON give "Unexpected token" and where is the error?

Accepted Answer

The most common causes are a comma after the last item, single quotes, an unquoted key, a comment, and a missing comma between two values. JSON.parse reports only the first problem and, in some browsers, only a character position. This formatter lists every error in the text, each with line and column, shows the snippet with an arrow under the exact spot, marks the line in the editor margin, and has a Go to the error button that selects the spot.

Question 2

Can JSON have comments or a trailing comma?

Accepted Answer

No. RFC 8259 has neither comments nor trailing commas, and JSON.parse rejects both. Files such as tsconfig.json and VS Code settings use JSONC, a variant that accepts comments and trailing commas, and JSON5 goes further. If you paste one of those, Repair mode removes the comments and the extra commas and lists each removal so you can decide whether to apply it.

Question 3

Do single quotes work in JSON? What about a Python dict?

Accepted Answer

No, JSON accepts only double quotes. Printing a Python dict gives single quotes and writes True, False and None, which in JSON are true, false and null. Repair mode converts the quotes and those three values. For data going from Python to another system, prefer json.dumps, which produces real JSON.

Question 4

What does Repair mode do and what are its limits?

Accepted Answer

It fixes what has an obvious reading: commas, straight and curly single quotes, unquoted keys, comments, True/False/None, NaN and Infinity (they become null), numbers like .5 or +1, hexadecimal, Markdown fences, and brackets or strings left open by truncation. Numbers with a leading zero, like 007, become strings so the zeros are not lost. Several documents pasted one after another become an array. It guesses the intent and can be wrong in ambiguous cases, which is why it shows the list of changes before you apply anything. Errors with no possible reading, such as a stray word at the very start, are not repaired.

Question 5

What happens when the JSON has the same key twice?

Accepted Answer

RFC 8259 recommends unique names and says that the behavior of software receiving repeated names is unpredictable. In JavaScript, JSON.parse keeps only the last value and the rest vanishes without notice, which hides configuration bugs. Here the second occurrence gets a warning with the line of the first, the output keeps both, and the JSONPath query uses the last one, as JavaScript does.

Question 6

Why does a big number change after going through JavaScript?

Accepted Answer

JavaScript stores numbers as 64-bit doubles, which represent integers exactly only up to 9007199254740991 (2^53 - 1). Above that, neighboring values collapse into the same number: 9007199254740993 is read as 9007199254740992. Decimals with more than 15 to 17 significant digits are rounded too. The formatter warns about each case and keeps the digits in the output text. For 64-bit ids in APIs, the safe route is to carry them as strings.

Question 7

Does sorting keys or minifying change what the JSON means?

Accepted Answer

No. JSON objects are unordered collections, so reordering keys does not change the data, it only makes two files easier to compare. Arrays keep the order of their items. Minifying only removes whitespace and line breaks outside strings. Numbers and string escapes come out as they went in: \u00e9 stays \u00e9.

Question 8

How do I validate a JSON Lines or NDJSON file?

Accepted Answer

Turn on JSON Lines under More options (or open a .jsonl or .ndjson file, which turns it on by itself). Each line is read as an independent document and the error points to the line in the whole file, not the line inside the record. The output is one compact value per line, which is the JSON Lines format (UTF-8, one value per line). Blank lines are ignored here, although the format does not allow them. The Tree and the Query treat the file as an array of records.

Question 9

Which query syntax is supported?

Accepted Answer

A JSONPath subset inspired by RFC 9535 (2024): $ for the root, .name and ['name'], indexes and negative indexes such as [-1], the wildcard [*] and .*, unions [0,2], slices [1:3] and [::2], recursive descent ..name and ..[*], and a simple filter such as [?(@.price < 10)], [?(@.isbn)] or [?(!@.isbn)] with ==, !=, <, <=, > and >=. There is no && or || in filters and no functions. A query outside the supported syntax shows the error and its position, never a disguised empty result. Each result comes with its normalized path, and the limit is 5,000 results.

Question 10

Is my JSON sent to any server? What about large files?

Accepted Answer

No. Reading, formatting, the tree and the query all run in your browser, inside a Web Worker, a separate thread that keeps the page responsive. In an internal test with 6.6 MB and 40,000 records, reading took about 0.4 seconds and formatting with key sorting about 0.4 seconds, measured in Node. Upload accepts up to 10 MB, the tree shows children 100 at a time, and the maximum nesting accepted is 1000 levels.

JSON Formatter and Validator