kordoc

kordoc parses HWP 3.x and 5.x, HWPX, HWPML, PDF, XLS and XLSX, DOCX, PPTX and images (PNG, JPG, WebP) into Markdown and structured data. It is built for Korean public-sector documents, where HWP files are common and hard to handle with general tools, and the README says it was written by a local government employee who spent seven years wrestling with them. It can be used as a library, a CLI or an MCP server.

Beyond reading, it covers comparison and format-preserving patching of HWPX and HWP files, writing Markdown out to HWPX with official-document presets, filling forms and fields, and review features such as structure checks and personal-data masking. OCR runs locally. The README reports a composite 0.963 on the opendataloader-bench PDF set (200 documents, first among 12 public parsers, recorded 2026-09-29) and 100% table-structure match on 9,865 tables across 2,286 HWPX documents. Install with `npm install kordoc` or run `npx kordoc`; `npx -y kordoc setup` registers 17 MCP tools in Claude Desktop, Claude Code, Cursor and Codex. Java and Python SDKs and an offline install mode (`KORDOC_OFFLINE=1`) are documented.