markitdown

name	markitdown
description	Convert DOCX / PPTX / XLSX / HTML / EPUB to markdown via Microsoft markitdown. Triggers: markitdown, convert document, extract docx, extract pptx, extract xlsx, .docx file, .pptx file, .xlsx file.
effort	fast
context	inline

/markitdown — Document → Markdown extraction

Optional capability built on Microsoft markitdown. Wraps MarkItDown().convert() so the same flow works whether you call it from a skill, the CLI, or scripts.

When to use

• DOCX / PPTX / XLSX / HTML / EPUB / MOBI / audio / image — formats Claude Code's built-in Read tool doesn't handle
• Batch / non-interactive doc extraction (e.g. inside /batch plans)

When NOT to use

• PDF — prefer /pdf skill (uses Claude Code's built-in Read which has tuned PDF support and pages parameter)
• Plain text / markdown / code files — just use Read

Install (opt-in)

TAUSIK convention #19: zero external deps. markitdown is not in the bootstrap manifest. Install on demand:

pip install 'markitdown[all]'           # all converters incl. audio/image OCR
pip install 'markitdown[docx,pptx,xlsx]' # narrower install

tausik doc extract reports a clear error if markitdown isn't available — it does not crash TAUSIK.

Usage

CLI

.tausik/tausik doc extract path/to/file.docx
.tausik/tausik doc extract slides.pptx --format=pptx

Output: markdown on stdout. Exit code 1 if markitdown is missing or extraction fails.

From scripts

import doc_extract

md = doc_extract.extract_to_markdown("path/to/file.xlsx")
if md is None:
    # markitdown missing, file missing, or extraction failed (diagnostic on stderr)
    fall_back_to_something_else()
else:
    print(md)

extract_to_markdown() never raises — callers can rely on None for graceful degradation.

Operations

Command	Action
.tausik/tausik doc extract <file>	Convert to markdown
.tausik/tausik doc extract <file> --format=X	With format hint (logged, markitdown auto-detects)
python -c "import doc_extract; print(doc_extract.is_available())"	Check if markitdown is installed

Gotchas

• Audio / image conversion needs extras. pip install 'markitdown[all]' pulls in heavy deps (Azure cognitive services, OCR). Use [docx,pptx,xlsx] for office-only.
• Large XLSX files can produce huge markdown. Consider head -n 200 before piping to context.
• HTML extraction loses some structure (script/style stripped). Use Read for small HTML files.
• Reserved future use: brain_post_webfetch hook may eventually use doc_extract for HTML pages — not implemented yet.