Anthropic’s Claude (Sonnet, Opus, Haiku) emits clean GitHub-Flavored Markdown — the kind our converter was built for. Copy from the chat, paste, export DOCX. Code blocks, tables, citations, and LaTeX equations all carry over.
docx-package fallback always available.Hover over any Claude response and click the copy icon in the action menu (it sits next to the regenerate and rate buttons). The button copies the underlying markdown — code fences, tables, LaTeX, and lists all intact.
Avoid manual text selection. Claude renders markdown to styled HTML in the chat interface; selecting and copying the rendered output strips the underlying syntax. The copy icon gives you the raw markdown the model actually produced.
Visit the homepage. Markdown Input is the default mode — leave it on. The left panel is where Claude's output goes; the right panel renders a paper-clean preview as you type.
The converter is a single-page app that loads in well under a second. No signup, no account, nothing to install. Works on any modern browser including the one you already have open.
Cmd-V (Mac) or Ctrl-V (Windows/Linux). Headings, ordered/unordered lists, code blocks, tables, blockquotes, and KaTeX equations all render live in the preview pane.
If the preview looks wrong, the markdown is wrong. Fixing it in the editor is faster than fixing it in Word later. Check that fenced code blocks have language hints (```python) and that LaTeX uses dollar-sign delimiters.
Click DOCX in the export bar. Behind the scenes the markdown POSTs to /api/convert-to-docx (server-side Pandoc) for highest fidelity — better tables, native Office Math equations, cleaner typography. The .docx downloads instantly and opens in Word, Google Docs, Pages, or LibreOffice.
If the Pandoc route is unreachable, the converter falls back to a fully-local DOCX path (the `docx` npm package). Both produce a valid Word document; Pandoc just wins on edge cases like nested tables, footnotes, and OMML equation structure.
Claude’s markdown output has a few patterns worth flagging — none break the export, but knowing them helps you predict what lands in Word.
When Claude cites sources, it sometimes uses markdown footnotes (`[^1]`). The Pandoc DOCX path handles these natively. The fully-local fallback drops them — if footnotes are critical, make sure /api/convert-to-docx is reachable, or convert via PDF/HTML first.
Code Claude produces in artifacts (the right-side panel) renders identically to inline code blocks once copied. Use the artifact's copy button to grab the raw markdown — copying from the rendered code view loses the language hint.
Claude's longer outputs (5,000+ words with mixed code, tables, and prose) export cleanly. The PDF path handles smart page breaks across all block elements; DOCX preserves heading hierarchy for navigation pane support.
Claude tables use GFM pipe syntax. The Pandoc DOCX export converts them to native Word tables (real cell grid, alignment honored). The fully-local DOCX fallback simplifies tables to text rows.
Inline (`$x$`) and block (`$$ ... $$`) both work. DOCX exports translate equations to native Office Math (OMML) when going through Pandoc — fully editable in Word. PDF and HTML render via KaTeX.
When Claude shows tool calls or function inputs/outputs (function_calls, tool_use), they render as code blocks in the export. They look clean but lose the structural distinction from regular code.
For export purposes, no. All three Claude variants emit identical GFM markdown structures. The differences are in what they generate — pick by use case, not by export compatibility.
Daily driver for most documentation work. Strong at structured markdown with consistent heading hierarchy.
Long-form writing, technical specs, research summaries. Tends to produce richer formatting (more nested lists, tables, callouts).
Fast turnaround on shorter content. Markdown output is identical in structure — just less of it.
Copy any Claude reply, paste, and you’ll have a DOCX in seconds.
Open the converter