&, < and > are escaped as HTML entities in metadata values

>>> metadata = cli.request('CreateMetaData', id=ELEMENT_ID, body={
    "type": "text",  # Any type will work
    "name": "some name",
    "value": "&<>",
})
>>> print(metadata['value'])
&amp;&lt;&gt;

This is probably related to the HTML sanitization that gets executed for markdown metadata; it should only run for markdown metadata though, not every single metadata type.