Work with the original document
Continue in the shared XML workspace with source and element outline. Back to website keeps it mounted; Resume editor returns to it. Browser sessions are temporary, so download work you want to keep.
Free · local XML tools
Convert XML to ordered JSON
Choose XML, review an ordered representation of its parsed nodes, then download JSON. Your original XML stays available beside the conversion tool.
Step 1: Choose your XML
Original XML stays in your browser. Task inputs: up to 8 MiB combined.
Paste XML instead
Output uses an ordered XML node representation: attributes and children stay distinct; values remain strings.
- Step 2: Configure and review output
Your original XML opens once in the existing editor with Convert XML selected. Choose Review converted copy; CSV also needs explicit node selectors for each column.
- Step 3: Review and keep a copy
Review warnings and the output preview, then download a separate JSON or CSV copy. The original XML is retained; edits or mapping changes require a new review.
Explicit mapping and output limits
JSON keeps parsed node order and namespace-aware names. CSV may flatten markup and cannot distinguish missing and empty strings. Conversions are limited to 8 MiB decoded UTF-8 input/output, 10,000 CSV rows and a 40-second cancellable local worker. No external XML resources are loaded.
Frequently asked questions
What JSON structure is created?
The output has document and element nodes, namespace-aware names, ordered attribute arrays and ordered children. Text, CDATA, comments and processing instructions stay distinct. Repeated children stay separate nodes in the ordered array; this is not an automatic field-name projection.
Are numbers and XML formatting preserved?
Values remain strings, including identifiers and integers beyond JavaScript’s safe range. XML declaration, original encoding, entity spellings and quote choices are not a lexical round trip. Keep the original XML when those details matter.
What are the limits and privacy rules?
Conversion runs in a terminable local worker with Cancel and a 40-second timeout. Input and output each allow 8 MiB of decoded UTF-8 text, with depth, token and node budgets. DTDs and external resources are refused. No account or AI provider is required; ordinary hosting and analytics requests are separate.