doc-to-markdown（DOC TO Markdown 控制）

Doc-to-Markdown (Word → Markdown)

Convert Microsoft Word .doc / .docx into:

• a clean Markdown file (.md)
• plus an optional images folder (*_images/) with relative image links

This is designed to keep Markdown small (good for humans + LLMs) while preserving diagrams.

Quickstart (copy/paste)

# 1) Convert a single file (.docx or .doc)
python3 convert_word_to_markdown.py "path/to/document.docx"

# 2) Embedded mode (single self-contained .md, very large)
python3 convert_word_to_markdown.py --embedded "path/to/document.docx"

# 3) If anything fails, run a dependency check
python3 convert_word_to_markdown.py --check

Batch convert (current folder)

for f in *.doc *.docx; do
  [ -e "$f" ] || continue
  python3 convert_word_to_markdown.py "$f"
done

Outputs

Default (external images):

document.docx
document.md
document_images/
  image1.png
  image2.png
  ...

Embedded mode:

document.docx
document.md   # contains base64 images

Requirements

• Recommended (most reliable): install markitdown into a local virtualenv in this repo

- bash setup_venv.sh - (manual) python3.11 -m venv.venv + .venv/bin/python -m pip install 'markitdown[all]'

• Alternative: install markitdown globally

- python3 -m pip install 'markitdown[all]' (requires Python 3.10+ and markitdown on PATH)

• Fallback: uv (provides uvx) so the scripts can run markitdown without pip installs

- macOS: brew install uv

• For .doc (legacy) support: LibreOffice (brew install --cask libreoffice)

Environment Overrides (for reliability)

• MARKITDOWN_UVX_PYTHON=3.11 (default) — change the Python version used by uvx
• MARKITDOWN_UVX_OFFLINE=0 — allow uvx to use network (default: offline)
• MARKITDOWN_CMD="... markitdown" — full command override (advanced)
• UV_CACHE_DIR=/tmp/uv-cache — use this if uvx can’t write to its cache directory (default: ./.uv-cache/)

Common Failure Modes

• .doc conversion fails:

- LibreOffice GUI running → quit LibreOffice (or killall soffice) and retry - If you see Abort trap: 6 / exit 134 in a sandboxed tool runner → pre-convert .doc to .docx outside the sandbox, then convert the .docx

• WMF/EMF diagrams don’t display: in sandboxed environments the WMF/EMF → PNG step may be skipped; convert those images to PNG outside the sandbox if needed
• markitdown not found: create ./.venv/ (recommended) or install markitdown globally
• Failed to initialize cache at ~/.cache/uv: set UV_CACHE_DIR=/tmp/uv-cache and retry

Notes

• convert_word_to_markdown.py is the entrypoint (handles both .doc and .docx).
• convert_with_images.py is an internal helper and only supports .docx.