SDS to XML Converter API

The structure, by example

This is the XML for a small record (indented here for reading; the API returns it without whitespace between elements). It is the same data as outputs.json, with confidence_score and warnings added as final elements.

<?xml version="1.0" encoding="UTF-8"?>
<sds_extraction>
  <product_identification>
    <product_name>Acetone</product_name>
    <supplier_name>Example Chemicals</supplier_name>
  </product_identification>
  <hazards_identification>
    <ghs_classification>Flammable Liquid - Category 2</ghs_classification>
    <ghs_classification>Eye Irritation - Category 2A</ghs_classification>
    <signal_word>Danger</signal_word>
  </hazards_identification>
  <composition>
    <substance_name>Acetone</substance_name>
    <cas_number>67-64-1</cas_number>
    <concentration_percent>95-100</concentration_percent>
  </composition>
  <physical_chemical_properties>
    <flash_point_c>-20</flash_point_c>
    <odor></odor>
  </physical_chemical_properties>
  <transport_information>
    <un_number>UN1090</un_number>
    <hazard_class>3</hazard_class>
    <packing_group>II</packing_group>
  </transport_information>
  <confidence_score>0.94</confidence_score>
  <warnings>Section 9: boiling point unreadable</warnings>
</sds_extraction>

The rules

• The root element is sds_extraction, and element names are the JSON keys, in the same order.
• An array becomes repeated elements with the same name: two ghs_classification values give two ghs_classification elements.
• An array of objects repeats the parent: each substance is its own composition element containing its fields.
• A null or empty string becomes an empty element, such as <odor></odor>, and an empty array becomes one empty element.
• There are no attributes and no namespaces. Values are XML-escaped (& < > and quotes).
• Tag names are sanitized: spaces become underscores, characters outside letters, digits, _, . and - become underscores, and a name that does not start with a letter gets a leading underscore.
• No XSD or DTD is published. The contract is the set of 16 section keys; fields inside a section follow the document.

Consequences to plan for

• One item looks like a scalar. A list with one entry is a single element, the same as a plain value. Always read repeated elements with a find-all, never assume a count.
• Empty is ambiguous. An empty element can mean null, an empty string or an empty array. If the difference matters, use the JSON output.
• Confidence and warnings are in the document. In XML and CSV they are appended to the record. In the JSON response they are top-level fields beside outputs.

Parsing it

Python's standard library is enough. This snippet runs against a real response:

import xml.etree.ElementTree as ET

root = ET.fromstring(result["outputs"]["xml"])   # result = client.extract("sds.pdf")

un_number = root.findtext("transport_information/un_number")
confidence = float(root.findtext("confidence_score"))
warnings = [w.text for w in root.findall("warnings") if w.text]

# composition repeats once per substance, and a single substance is still one <composition>
substances = [{child.tag: child.text for child in item} for item in root.findall("composition")]

XML, JSON or CSV?

How CSV is written

field,value
product_identification.product_name,Acetone
hazards_identification.ghs_classification,"[""Flammable Liquid - Category 2"",""Eye Irritation - Category 2A""]"
composition[0].substance_name,Acetone
composition[0].cas_number,67-64-1
physical_chemical_properties.flash_point_c,'-20
physical_chemical_properties.odor,
transport_information.un_number,UN1090
confidence_score,0.94
warnings,"[""Section 9: boiling point unreadable""]"

• The header is field,value, with one row per value and dotted paths such as transport_information.un_number.
• An array of objects is indexed, for example composition[0].cas_number. An array of plain values is written as a JSON string.
• A value that starts with =, +, - or @ gets a leading apostrophe, which stops spreadsheets running it as a formula. That is why -20 appears as '-20: strip a leading apostrophe when you import numbers.

How you get access: the free sample upload on the homepage needs no key, the self-serve plan issues an API key after signup, and enterprise customers get a dedicated deployment with its own base URL. See pricing.

FAQ

Related pages

• MSDS to JSON
• SDS field coverage matrix
• OpenAPI contract
• Python and JavaScript clients