kordoc, an Open-Source MCP Server for Converting Korean Documents to Markdown, Released

kordoc is a CLI tool and MCP (Model Context Protocol) server that converts HWP 3.x/5.x, HWPX, HWPML, PDF, XLS/XLSX, DOCX, and images (PNG/JPG/WebP) to Markdown. Written in TypeScript, it is distributed as an npm package and runs on Node.js 18 or higher.

Key features:

• RAG pipeline support: The --format chunks option outputs structured chunks (JSON) that preserve heading hierarchy as breadcrumbs
• Local OCR: Built-in Korean PP-OCRv5 model (~18MB) enables scanned document processing without an API key
• Formatting-preserving roundtrip: When edited Markdown is applied back to the original HWPX/HWP, only the text is replaced while all other formatting is preserved. In its own benchmark, it achieved 99.998% HWPX text reproduction accuracy and 100% table structure accuracy (across 1,421 tables)
• Markdown → HWPX generation: Supports standard administrative document formats (presets for draft documents, reports, and announcements)

When registered as an MCP server, AI clients such as Claude Desktop, Cursor, and Claude Code can directly call a total of 15 tools, including parse_document, fill_form, and redact_document. A single command, npx -y kordoc setup, automatically detects installed AI clients and patches their configuration files.

It is released under the MIT license, allowing commercial use.