# kordoc [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/chrisryugj/kordoc  
**GitHub Stars:** 2542  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/kordoc

## Description
Parse Korean documents (HWP, HWPX, PDF, XLSX, DOCX, PPTX) to Markdown; fill forms, diff, render.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What kordoc MCP server does

The kordoc MCP server exposes document processing capabilities to MCP-compatible AI clients. It accepts local HWP 3.x and 5.x, HWPX, HWPML, PDF, XLS/XLSX, DOCX, PPTX, PNG, JPG, and WebP files. Parsed output can be returned as Markdown, structured intermediate data, page-level content, or chunks designed for retrieval-augmented generation.

The toolset also covers document operations beyond extraction. Agents can compare documents at block and cell level, apply edits while preserving HWP or HWPX formatting, create HWPX files from Markdown, populate fields and form controls, apply seals to supported forms, render or preview documents, extract regions, run structure and notation checks, and mask personal information. The README describes 17 MCP tools spanning parsing, comparison, generation, and rendering.

## How it works

kordoc reads files directly rather than requiring Hancom Office. For scanned documents and image content, it can use local OCR. OCR output can include line boxes and confidence values when the corresponding options are enabled. The command-line and API examples show the same underlying processing model: parse a file, inspect Markdown and structured blocks, or generate an HWPX document from Markdown and configuration.

For retrieval workflows, the parser can attach heading paths to chunks and emit tables as independent chunks. Tables can be represented in GitHub-Flavored Markdown instead of HTML. Formatting-preserving edits use Markdown extracted with the layout-table option; edits that cannot be applied are skipped and reported with a reason.

## Setup and configuration

Install Node.js 20 or later, then install the package with `npm install kordoc`. The CLI can also be run without a permanent installation through `npx kordoc`. PDF and OCR dependencies are installed by default; omitting optional dependencies limits those functions.

Run `npx -y kordoc setup` to register the kordoc MCP server with installed Claude Desktop, Claude Code, Cursor, and Codex clients. Restart the selected client after registration. A Claude Code plugin is also available through its marketplace commands. Manual MCP registration and troubleshooting are documented by the project.

For restricted environments, set `KORDOC_OFFLINE=1` to block external communication. OCR can run locally on CPU without an API key. `KORDOC_ROOT` is available to define the MCP access scope. The project also documents Python and Java SDKs that call a resident worker, but those are separate from the MCP client setup.

## Tools and capabilities

Supported workflows include:

- Parse office documents and images into Markdown, structured data, page content, or retrieval chunks.
- Compare blocks and spreadsheet-like cells across documents.
- Patch HWP and HWPX files while retaining supported formatting.
- Generate HWPX from Markdown with tables, formulas, charts, and government-document presets.
- Fill fields and form controls, including two standard draft-document templates.
- Preview, render, inspect regions, validate structure, and mask personal information.
- Run local OCR, including OCR for text inside HWP or HWPX images.

## Limitations and notes

The strongest validation results in the README concern Korean document tables and specific benchmark corpora; they should not be read as guarantees for every document. PDF and OCR accuracy varies by source, layout, fonts, and scan quality. The project separately records known rendering and parsing limitations, including cases involving unsupported fonts and HTML-line output.

Parallel batch conversion increases memory usage because files are processed concurrently. Formatting patches are not guaranteed to apply every edit: unsupported changes are omitted and reported. The offline mode blocks external communication, but PDF and OCR functionality still depends on the relevant packages being installed. The kordoc MCP server is therefore best suited to local-file workflows where the client can provide accessible paths and the required optional dependencies are present.

_Full upstream README: https://allmcps.com/mcp/kordoc/readme_