JSON repair

About this tool

Repair a copied configuration, API sample or generated JSON fragment in a local Web Worker using jsonrepair 3.15.0. The tool can add missing punctuation, quote keys, normalize single quotes and remove comments or trailing commas. It retains the original input and repaired text for comparison and shows one exact enclosing changed span. Repairs are guesses about syntax: a parseable result does not establish the intended data, restore missing content or validate a schema. Review every result before copying or downloading it.

Executable input and output examples

{
  // copied config
  "name": "Ada",
  "enabled": true,
}

{
  
  "name": "Ada",
  "enabled": true
}

{name: 'Ada', city: '東京'}

{"name": "Ada", "city": "東京"}

{"a": 1 "b": 2}

{"a": 1, "b": 2}

{"user":{"name":"Ada","roles":["reader","editor"

{"user":{"name":"Ada","roles":["reader","editor"]}}

{"message":"Hello

{"message":"Hello"}

{id: 9007199254740993, price: 1.2300, small: 1e-400, sign: -0,}

{"id": 9007199254740993, "price": 1.2300, "small": 1e-400, "sign": -0}

{
  "id": 9007199254740993,
  "price": 1.2300,
  "small": 1e-400,
  "sign": -0,
  "escaped": "\u0061"
}

{
  "id": 9007199254740993,
  "price": 1.2300,
  "small": 1e-400,
  "sign": -0,
  "escaped": "\u0061"
}

```json
{"ready": true,}
```

{"ready": true}

callback({"ok":true});

{"ok":true}

{'active': True, 'deleted': False, 'value': None}

{"active": true, "deleted": false, "value": null}

{"id":1}
{"id":2}

[
{"id":1},
{"id":2}
]

{"state": pending}

{"state": "pending"}

{"emoji":"😀", unquoted: true,}

{"emoji":"😀", "unquoted": true}

{"a":1,"\u0061":2}

{
  "error": "duplicate"
}

{"value":1e}

{
  "error": "numericChange"
}

{"value":}

{"value":null}

[1,,2]

[
[1],2
]

Common errors and review points

• A duplicate error can be caused by escaped spellings of the same key. Remove or rename the conflicting property after checking which value belongs there.
• A numeric-change error is intentional protection. Correct malformed numeric literals from their source rather than assuming a guessed exponent or decimal is right.
• An empty-input error also applies to whitespace-only text. Paste a JSON value or a JSON-like fragment first.
• Byte limits count UTF-8 bytes, not visible characters. Chinese or Japanese text and emoji can reach the cap sooner than ASCII.
• A depth error can occur before repair finishes. Reduce excessive nesting instead of repeatedly retrying the same document.
• A syntax error means the repair library could not produce an accepted result. Inspect unmatched delimiters, invalid escapes and unexpected surrounding text.
• A successful repair of a cut-off string is not proof that the string is complete. Compare it with the original source.
• A wide change span can cover several separate corrections and unchanged interior text. It is not a count of edits.
• If a task times out or you cancel it, no result is available to export. Try a smaller fragment and review any omitted context.
• Raw malformed Unicode cannot be exported faithfully as UTF-8. Correct it at the source; do not confuse it with a literal backslash-u escape.

Limits and notes

• —Input is capped at 200,000 UTF-8 bytes, output at 500,000 UTF-8 bytes and nesting at 64 levels. The browser terminates a repair worker after 5 seconds; this is not a streaming or large-file tool. Raw unpaired UTF-16 surrogates are rejected because a UTF-8 download would replace them. An already-escaped sequence such as "\uD800" is retained as text; downstream Unicode handling may differ.
• —Syntax repair cannot recover omitted records, infer business meaning or prove an API payload is correct. A closed truncated string can still contain incomplete data. This is not a complete JSON5 parser, schema validator or general JavaScript interpreter. Some malformed inputs remain unsupported and return no repaired output. Unknown or incomplete escapes, undefined, ellipses and JavaScript expressions such as [2/3], [true&&false] or {x:true?false:null} are outside the supported repair profile. A missing value may be filled with null; this is a guessed placeholder, not recovered data.
• —Duplicate decoded object keys are rejected, including "a" and "\u0061" in the same object. The tool does not silently retain the last value; the same key in separate objects is allowed. Numeric token spellings are checked before and after repair. A proposed numeric change, insertion or removal is rejected; examples include 01, 0x10, .5, 1. and an incomplete 1e. Valid numeric spellings such as 9007199254740993, 1.2300, 1e-400 and -0 are retained as text. A later JavaScript parser or another formatter can still round or normalize them.
• —The displayed difference is a single minimal enclosing changed span based on the common prefix and suffix, not a list of independent edits or a semantic JSON diff. Unchanged text inside that span is included.
• —JSONP removal, Python constant conversion, code-fence removal, NDJSON wrapping and bare-word quoting can change how the text is interpreted. These are proposals to review, not evidence of the author’s intent. The tool does not execute JavaScript, evaluate JSONP callbacks, resolve URLs, fetch data or store the input in browser storage. Copy and download happen only when requested.

Frequently asked questions

Related tools