This is the cleanest of the MBOX operations because it is essentially unpacking a container into its contents, one message becoming one EML and the routes run from script to tool:

The script route, again the most control: Python's mailbox module opens the MBOX, iterates messages and writes each with as_bytes to a numbered or subject named .eml file, a dozen lines producing the exact per message EML the ingesting system wants. The control matters here for naming, since the target system often has filename expectations, date prefixed or ID based, that a script satisfies precisely while a tool may not.

The client route: Thunderbird with ImportExportTools NG imports the MBOX and then exports the folder as individual EML files, the add-on's export all messages as EML operation doing exactly this split with a folder of results.

The converter route: MBOX to EML converters do the split as their whole purpose, keeping folder structure if the MBOX came from a multi folder export, useful when the source was a full mailbox rather than one folder.

The naming and encoding details that decide whether the ingesting system accepts the output: filenames need to match what the system expects and avoid characters it rejects, subject based names sanitized of slashes and colons and the EML output should preserve the original headers and encoding intact rather than re encoding, which a straight write of the message bytes achieves and a careless conversion can corrupt. Test one EML through the ingesting system before splitting the whole MBOX, the single file that validates the naming and format before you commit thousands.