PDF to Markdown: local conversion, cloud OCR, and the trade-offs

Getting text out of a PDF is easy to demonstrate and surprisingly difficult to get right.

A document can look perfectly readable while producing broken paragraphs, scrambled columns, or a table that turns into a wall of text. Scanned PDFs add another problem: there may be no text layer to extract at all.

I'm Leo, the maker of PDF to MD. It converts documents into editable Markdown, with local browser processing for text documents and a separate cloud OCR mode for scans.

Here's how the workflow works, and where it still needs human review.

Start by distinguishing text PDFs from scans

Before choosing a conversion method, try selecting a sentence in your PDF.

If you can select individual words, the document probably has a text layer. Local extraction is a reasonable starting point, though an existing text layer doesn't guarantee correct reading order.

If each page behaves like a single image, you'll generally need OCR.

Some PDFs contain both types of pages, so it's worth checking more than the first page.

Local conversion for text documents

PDF to MD can convert text-based PDFs, DOCX files, and HTML locally in the browser, without uploading file contents.

The workflow is:

1. Choose your documents.
2. Convert them to Markdown.
3. Review and edit the output.
4. Copy the Markdown, download individual files, or export a batch as a ZIP.

Local conversion is free. It's useful when you want to turn a document into notes, documentation, or text you can work with in another tool.

For example, an Obsidian workflow is to download the .md file, place it in your vault, and add your own links and annotations. This is a file-based workflow; PDF to MD is a web app, not an Obsidian plugin.

Cloud OCR for scanned documents

Scanned PDFs and images need a different processing path.

PDF to MD offers an optional cloud OCR mode for those files. Unlike local conversion, this mode uploads the document for processing.

The distinction matters: a tool offering local processing does not mean every feature runs locally. Choose the mode based on both the document type and whether you're comfortable uploading it.

Cloud features use account quotas, with paid plans and OCR packs available for additional usage.

Markdown still needs a review step

A PDF describes a page layout. Markdown describes document structure. Converting between them involves interpretation.

These are the areas I would check before using the output:

• Reading order: Do paragraphs from separate columns appear in the right sequence?
• Headings: Did section titles become headings, or just isolated lines?
• Tables: Are values still associated with the correct row and column?
• Formulas: Are symbols and relationships preserved accurately?
• Repeated content: Did page headers and footers end up inside the body?

Complex tables, formulas, and multi-column layouts can still need manual correction. Local PDF conversion also does not extract embedded images.

A Markdown file that renders successfully can still contain extraction errors. For technical or research material, compare important passages against the original.

Using the output in automated workflows

For repeated document processing, PDF to MD also provides a REST API and a remote MCP endpoint.

These offer another way to access conversion from scripts and compatible AI tools. They are separate from the free local browser workflow and use service quotas.

You can find the integration details in the developer documentation.

Whether the output goes into a notes app, documentation system, or AI workflow, checking the extracted text first is still worthwhile. A downstream tool cannot reliably recover a table whose columns were mixed up during conversion.

Try it with a document you know well

The easiest way to evaluate a converter is with a document whose contents you already understand. Start with a short text PDF, inspect the Markdown, and then try a more demanding layout.

You can try PDF to MD here. Free local conversion is available alongside optional cloud features.

I'd appreciate feedback from people moving documents into Markdown: which causes you the most cleanup work—tables, reading order, scanned pages, or something else?