Ask HN: How should I convert Microsoft Word documents to Markdown? | Hacker News Reader

Ask HN: How should I convert Microsoft Word documents to Markdown?

5 pointslkrubner8 comments

I took over a project that was built by an overseas team. They set up a data ingestion process. They have a step in the ingestion where they use Libre Office (in headless mode) to convert Microsoft Word documents to PDFs. Later we convert all PDFs to Markdown. They felt that it was best to convert everything to a PDF, and then convert all of the PDFs to Markdown.

What I notice is that LibreOffice can create very complex PDFs when the Microsoft Word document has:

1. tables

2. multiple columns

3. strikethrough text

I am thinking we should go straight from Microsoft Word to Markdown.

What is the right software for that?

Pandoc is awesome for this: `pandoc input.docx -o output.md` There's more you can do, with style sheets and so on, which you will likely have to dig into for the tables and multiple columns to come out the way you want. You can also extract media files from inside a docx file: `pandoc --extract-media=. input.docx -o output.md`

Native support: https://techcommunity.microsoft.com/blog/onedriveblog/introd...

Microsoft OSS python: https://github.com/microsoft/markitdown

There seem to be many addons that enable this, and pandoc as others have suggested

I'd give an llm a shot before I ruled it out.

I had it generating .docx the other day and it did pretty well, so I assume it understands the format just fine.

And they're excellent at markdown.

sure but OOXML is on avg 12x~ larger than markdown so you'll burn a ton of tokens. I second using pandoc but depends on why you need it to be in markdown in the first place?

(source - I'm working on a docx MCP that abstracts OOXML complexity away from LLMs entirely)

Start with pandoc before reinventing the wheel.

Pandoc might be able to do this, found this:

https://gist.github.com/plembo/409a8d7b1bae66622dbcd26337bbb...

Claude Code can do these types of really well unless you're trying to convert in bulk

Pandoc can do this I think

Reply on news.ycombinator.com