JSON to CSV converter, nested JSON and JSON Lines too

When an object holds several arrays of records, the one with the most records-like items is read and the note names the others it left out. Anything else, such as a single object with no array in it, opens as a table of one row.

Nested objects, arrays and missing fields

• Nested objects are flattened into columns named by their path, with dots between the names:{"reading": {"wind": {"speed": 8.7}}} gives the column reading.wind.speed.
• Arrays inside a record stay whole, as JSON text in one cell: ["ice"], written to the CSV as "[""ice""]" (a CSV doubles the quotes inside a quoted field). They are not spread over extra rows or numbered columns, so every record stays exactly one row.
• Missing fields and null give an empty cell. The columns are every key of any record, in the order they first appear, so a field that only some records have still gets its column.
• Values: numbers with 15 significant digits, true and false as TRUE and FALSE, text as it is, and ISO times such as 2025-01-15T09:00:00Z as 2025-01-15 09:00:00 (UTC, as in the file).

A worked example you can check

“Try a sample JSON file” opens 72 made-up hourly readings from Leeds, Cardiff and Belfast on 15 January 2025. Each record nests the station and the reading, has an alerts array, and only rainy hours have a rainfall. The first record:

{"time": "2025-01-15T00:00:00Z", "station": {"city": "Leeds", "code": "LDS"}, "reading": {"temperature": 2.0, "humidity": 88, "wind": {"speed": 8.7, "direction": "NE"}, "rainfall": 0.8}, "raining": true, "alerts": []}

Click Download as CSV, and the CSV, with its 10 columns, starts like this:

time,station.city,station.code,reading.temperature,reading.humidity,reading.wind.speed,reading.wind.direction,reading.rainfall,raining,alerts
2025-01-15 00:00:00,Leeds,LDS,2,88,8.7,NE,0.8,TRUE,[]
2025-01-15 00:00:00,Cardiff,CDF,4.3,94,8.2,W,,FALSE,[]

The Leeds record an hour later, with an ice alert and no rainfall, becomes:

2025-01-15 01:00:00,Leeds,LDS,0.6,89,13.9,S,,FALSE,"[""ice""]"

One reading has a null temperature and three have no wind direction: their cells are empty. A test opens the sample on every change to the site and compares its CSV with the file we expect, byte for byte, and does the same with the records written as JSON Lines. Download the sample file (JSON, 15 KB).

The same in Python

pandas’ json_normalize flattens nested objects the same way. On the sample file:

import json, pandas as pd

records = json.load(open("megarows-sample-uk-weather-readings.json"))
table = pd.json_normalize(records)
print(len(table), "rows,", len(table.columns), "columns")
print(table.to_csv(index=False).splitlines()[4])

It prints:

72 rows, 10 columns
2025-01-15T01:00:00Z,False,['ice'],Leeds,LDS,0.6,89,13.9,S,

The same 72 rows and 10 columns, in another order. pandas keeps the time as the text of the file and writes arrays the Python way (['ice']), where the CSV from this page has JSON (["ice"]), which other programs can read back. We ran it on 9 October 2026 with numpy 2.1.3, pandas 2.2.3, Python 3.12.15.

Private by design: nothing is uploaded

API exports and logs can hold customer or test data. This tool reads your file and writes the CSV in your browser, so the file and its name never leave your device. Big exports open too: JSON Lines and arrays of records have no fixed size limit, and the practical one is your device’s memory.