Overview

PubAX offers a set of open‑source AI tools for the public sector, built so the AI can read Korean legal, official and government documents exactly as they are published. The central product, kordoc, is shipped as an npm package that can run either as a command‑line interface or as an MCP (minimal container‑provisioning) server. Once installed with `npm install kordoc`, each invocation can take a single file or a folder of files and outputs Markdown files that retain the original formatting. The Markdown includes a structure that can later be queried or passed to other AI assistance tools.

Kordoc accepts Korean documents in many native formats: HWP 3.x and 5.x, HWPX, PDF, XLSX, DOCX, PPTX, and raster images. For instance, a user holds a batch of 200 government PDF reports; running `kordoc --input reports.zip` produces a `reports.md` that contains every heading, paragraph, table and figure in the same order as the source. The tool preserves every table: from 2,286 HWPX files containing 9,865 tables, the output Markdown reproduces those table boundaries and cell content without disappearance. The input files are treated as read‑only; no data is sent to the cloud, making the tool suitable for networks that cannot connect to external services.

The application also supports form‑filling: a user can supply a JSON payload that matches field names in a HWP or PDF form, and kordoc will replace the form fields with the supplied values. After finishing, the tool can generate a diff between the original file and the modified file, outputting a third Markdown file that lists added or changed content line by line. This is useful for auditors who need to verify that content changes are deliberate. The diff feature works on all supported formats, not just HWPX.

Kordoc is part of a larger suite that includes other MCP servers such as Korean law, statistics, patent, building, and school data. These MCP servers can be summoned by their URLs; for example, the law server is hosted at `mcp.gomdori.app/law`. Your AI assistant can then query that server directly for legal references, ensuring that any citation generated by the assistant can be cross‑checked against official sources. The MCP format allows lightweight deployment, and the server can run on a local machine or within an intranet, requiring no cloud API.