Convert PDF to Markdown

How this tool computes its result

Text extraction runs entirely client-side via pdfjs-dist. For each page it pulls textContent.items, buckets each text fragment by its rounded y-coordinate (transform[5]) to approximate a "line," sorts fragments within a line by x-coordinate left to right, and joins lines top to bottom by sorting line keys descending (PDF y-coordinates increase upward). Pages are joined with a blank line between them, and runs of 3+ newlines are collapsed to 2. This is positional-heuristic extraction, not real layout/structure analysis — there is no heading, list, or table detection.

Worked example

A 2-page PDF with distinct paragraphs, uploaded as "invoice.pdf" under 2 MB: buildPdfMarkdown returns pageCount: 2 and a markdown string that is really just plain text reconstructed from x/y-sorted text fragments, joined per page with a blank line, downloadable as "invoice.md".

When not to use this tool

Scanned/image-only PDFs with no embedded text layer will produce an empty extraction, which the component reports as "No text was extracted from this PDF." File size is hard-capped at MAX_FILE_SIZE = 2 * 1024 * 1024 bytes (2 MB); anything larger is rejected before any parsing is attempted.

Common mistakes

• - Uploading a file that is not a .pdf (checked only by filename extension, not MIME sniffing) — rejected with "Please upload a PDF file."
• - Uploading a PDF over 2 MB — rejected with "File is too large. Max size is 2 MB." before conversion even starts.
• - Expecting real Markdown syntax (##, **bold**, tables) in the output — the tool only reconstructs plain text line order from coordinates; it does not infer or emit any Markdown formatting markers.

Related tools