Get the page

[Fianax]

I'm using markitdown to edit word.

But when I pull the text I don't know which page it belongs to, which is important to have traceability when I generate the chunks from that word.

Is there a way to get it with the current version?
If not, do you plan to do it in some future versions? because I see it interesting and useful.

[ltianyi992]

Page numbers are not accessible through markitdown currently, and this is a fundamental DOCX format limitation rather than a markitdown gap.

Why DOCX does not store fixed page numbers:
Word's .docx XML stores content as a semantic flow (paragraphs, runs, tables) — actual page breaks are calculated at render time by the Word rendering engine based on font, margin, and printer settings. The XML does contain <w:lastRenderedPageBreak> hints from the last time the document was saved, but these are unreliable artifacts, not authoritative page boundaries.

Workarounds for traceability:

Option 1 — Convert to PDF first, then extract with page metadata

# Use LibreOffice headless to produce a PDF, then pdfminer for page-aware extraction
import subprocess, pdfminer.high_level

subprocess.run(["soffice", "--headless", "--convert-to", "pdf", "doc.docx"])
pages = list(pdfminer.high_level.extract_pages("doc.pdf"))
for page_num, page in enumerate(pages, start=1):
    # extract text per page
    print(f"Page {page_num}: ...")

Option 2 — Use python-docx to detect explicit page breaks

from docx import Document
from docx.oxml.ns import qn

doc = Document("doc.docx")
page = 1
for para in doc.paragraphs:
    # Check for explicit page break in runs
    for run in para.runs:
        for br in run._r.findall(qn("w:br")):
            if br.get(qn("w:type")) == "page":
                page += 1
    text = para.text
    if text:
        print(f"[Page ~{page}] {text[:80]}")

This catches manually-inserted page breaks but not soft/automatic ones.

Option 3 — Embed page markers in the source document
If you control document creation, add bookmarks or custom XML properties at page boundaries and parse those as anchor points.

Future support for page metadata in markitdown would likely require an optional rendering pass (e.g., via LibreOffice UNO) — worth opening as a feature request if this is a common need for your use case.

[adarshajitsalehittal-code]

import subprocess
import pdfminer.high_level

Step 1: Convert DOCX → PDF

subprocess.run([
"soffice",
"--headless",
"--convert-to",
"pdf",
"doc.docx"
])

Step 2: Read PDF page by page

from pdfminer.high_level import extract_pages

pages = list(extract_pages("doc.pdf"))

for i, page in enumerate(pages, 1):
print(f"\n--- PAGE {i} ---")

✔ Accurate page mapping
✔ Works for any DOCX
✔ Industry standard approach

🥈 Lightweight method (no conversion)
👉 Use python-docx + manual page hints
from docx import Document
from docx.oxml.ns import qn

doc = Document("doc.docx")
page = 1

for para in doc.paragraphs:
for run in para.runs:
brs = run._r.findall(qn("w:br"))
for br in brs:
if br.get(qn("w:type")) == "page":
page += 1

print(f"[Page ~{page}] {para.text}")