Ask a coding agent to turn a Word file into a PDF and, with no LibreOffice on the machine, it usually falls back to pandoc — which rebuilds the document through LaTeX, drops the Word styling, and reports success anyway. This renders server-side and keeps the fonts the document asked for. For anything a browser can already do, it declines and names the free local package instead.
A client for the hushvert hosted conversion API. It handles the conversions a browser cannot do — office documents to PDF, PDF to Word, document interchange formats, and video transcodes — and writes the result next to the input file.
- convert_file — convert a local file and get the output path back
- convert_poll — finish a long conversion that was still running when convert_file returned, which is what a large video always needs
- list_formats — check a source-to-target pair is supported before the upload rather than after
- check_usage — free monthly conversions remaining, credit balance, and the current billing window
- Office to PDF (docx, pptx, xlsx, doc, ppt, xls, odt, ods, odp, rtf, html), PDF to Word, md/html/epub/latex/rst interchange, and mov/mkv/avi/webm to mp4
The npm package @hushvert/mcp over stdio, with HUSHVERT_API_KEY in the environment. HUSHVERT_ALLOWED_DIR confines every file read and write to one directory. Images, audio, archives and PDF page operations are refused — those run locally in the free @hushvert/engine package.
One command plus a key — claude mcp add hushvert -e HUSHVERT_API_KEY=hv_live_your_key_here -- npx -y @hushvert/mcp, then supply credentials