CSV to JSON Converter

What the JSON is usually for

CSV is what data arrives in. JSON is what code wants to be handed. Five jobs account for nearly every conversion I do in this direction:

• Test fixtures and mock responses. Someone sends 200 sample records in a spreadsheet. The frontend needs an array it can render against before the real endpoint exists.
• Seeding reference data. Country lists, tax rates, plan tiers, feature flags. Business people keep them in a sheet, the app reads them as JSON.
• Document store imports. MongoDB, Firestore, DynamoDB and Elasticsearch all ingest JSON. None of them takes a CSV without a conversion step first.
• Fine-tuning datasets. Training files for language models are JSON Lines, one example per line. Keep the prompt and completion columns in a sheet, switch Structure to JSON Lines, download the .jsonl.
• Anything spoken over HTTP. Dropping a JSON array into a request body or a fetch mock takes seconds. Dropping a CSV in there takes a parser.

JSON array or JSON Lines

The same five rows, with Structure switched to JSON Lines:

{"store":"Fenway","zip":"02215","units":48,"restock":true}
{"store":"Mission","zip":"94110","units":12,"restock":false}
{"store":"Lakeview","zip":60614,"units":0,"restock":true}
{"store":"Midtown","zip":30309,"units":7,"restock":false}
{"store":"Old City","zip":19106,"units":23,"restock":false}

• Pick the array when the whole thing has to be one valid JSON document: a request body, a config file, a fixture you import, anything a single JSON.parse has to swallow.
• Pick JSON Lines when something reads the file a record at a time. Log shippers, BigQuery and Snowflake loaders, and every language model fine-tuning format expect exactly this. A 2 GB file is readable line by line without parsing the rest.
• The formatting toggle applies to the array. Pretty gives two-space indentation, minified gives one long line. JSON Lines is always one compact object per line, because indenting it would break the one-record-per-line contract it exists for.
• The extension follows the choice. sales.csv comes back as sales.json or sales.jsonl, keeping the name you gave it.

Gotchas worth knowing

• Quotes and line breaks inside fields survive. A cell written as "He said ""hi"", then left" becomes one JSON string with ordinary quote characters in it. A cell containing a line break stays a single value, escaped the way JSON requires.
• A byte order mark is stripped. Files saved by Excel often start with an invisible marker. Left alone it becomes part of the first key name, which produces a bug that survives three rounds of debugging because the key looks correct on screen.
• Ragged rows are padded, not rejected. Rows are widened to match the longest one and the widget tells you how many it padded. If a data row is wider than the header, that extra field arrives under a generated name like column_5. Blank header cells get the same treatment.
• Duplicate header names collapse. Two columns both called id produce a single id key holding the value from the right-most one. Rename one of them before converting.
• Dotted headers do not create nesting. A column named user.name becomes a key spelled user.name. The output is one flat object per row, by design.
• 100 MB is the ceiling here. No row limit, no metering, but past 100 MB the browser tab is the wrong tool and the widget offers the full editor instead.