koul777/public-regulation-mcp-builder

PR MCP Builder is a Windows tool for turning official South Korean public institution regulation documents into a searchable knowledge base that AI programs can query safely. It takes files in PDF, HWP, HWPX, and DOCX formats and organizes them into a structure of institution, regulation, revision version, and chapter, section, article, so that when a revised version of a regulation is uploaded, the tool reads the enactment and amendment history at the top of the document and links it to the correct existing regulation series instead of treating it as unrelated. Rather than chopping every document into same sized chunks and relying only on similarity search, the tool first narrows down by regulation name and table of contents, then finds the article text in the most recently valid revision, tracing back through older revisions when needed. Crucially, only content that a human has reviewed and approved gets included in the search index or in answers given through MCP, which stands for Model Context Protocol, the standard this tool uses to expose the regulation data to AI assistants like ChatGPT, Claude, and Codex. The project includes a Streamlit based local interface for the person running it to process, review, and approve documents before a connection bundle is generated. It supports connecting to several different AI programs, including ChatGPT Desktop, ChatGPT's remote HTTPS mode, Claude Desktop, and Claude Code, each with its own setup steps and automated verification that the connection actually works end to end. Beyond basic parsing, the tool pays close attention to correctness for legal and administrative documents: it applies security level and department access rules so that higher clearance content is not shown to lower clearance viewers, it handles repealed or superseded regulations and retroactive amendments by defaulting to the safer option when unsure, and it works around known quirks in Korean HWP and HWPX table formats, DOCX merged table cells, and Unicode text normalization to keep search results accurate. The README states this is an actively developed, source available project, not yet a finished product, and warns that real deployment inside a government or public institution would need further validation of document formats, revision history handling, and security policy before use. It is released under the MIT license and requires Python 3.11 or newer on Windows 10 or 11.