sysprog21/zhtw-mcp

A linguistic linter for Traditional Chinese (zh-TW)

What it solves

AI language models often produce Traditional Chinese (zh-TW) text that contains "regional drift"—terminology, punctuation, and character shapes that follow Mainland Chinese (zh-CN) standards rather than Taiwan's Ministry of Education (MoE) standards. This project provides a linguistic linter to automatically detect and correct these inconsistencies to ensure text is authentic to how people in Taiwan actually write.

How it works

The tool operates as a linguistic linter that enforces three primary standards: punctuation marks (Revised Handbook of Punctuation), character shapes (Standard Form of National Characters), and cross-strait vocabulary normalization (based on OpenCC datasets).

It can be used as a standalone CLI tool or as a server via the Model Context Protocol (MCP), allowing AI coding assistants to call it as a tool. It features two strictness profiles (base and strict) and optional flags for software UI strings (relaxed) and AI writing artifact detection (detect_ai). For ambiguous terms, the server can query the AI assistant it is running within to help decide the correct term.

Who it’s for

Developers and writers who need to ensure their AI-generated Traditional Chinese text adheres to official Taiwan MoE standards, particularly in software localization and professional writing.

Highlights

  • MCP Integration: Plugs directly into AI assistants (like Claude Code or Codex) to provide real-time linting and auto-fixing.
  • Comprehensive Rule Set: Includes over 1,100 vocabulary rules and 15 casing rules.
  • MoE Compliance: Enforces official Taiwan Ministry of Education standards for punctuation and character variants.
  • AI-Aware Linting: Includes a detect_ai mode to flag filler phrases and semantic safety words common in AI writing.
  • Markdown Support: Can exclude fenced code blocks and HTML blocks from scanning to avoid breaking code.
  • Multi-platform Binaries: Provides pre-built binaries for Linux, macOS, and Windows.

Related

  • Project
    Raymondhou0917/speak-human-tw

    A proofreading skill for AI agents that removes "AI flavor" from Traditional Chinese text and localizes it for Taiwan, ensuring content feels human-written and natural.

  • Project
    KKKKhazix/human-writing

    A framework and toolset for AI agents to eliminate "AI-style" Chinese writing by enforcing human-like rhythms, factual substance, and structural checks.

  • Project
    larashero3-dotcom/lieflat-less-ai-tone

    A linguistic research project and toolset that quantifies "AI flavor" in Chinese text and provides a set of rules to remove these markers for more human-like writing.

  • Dispatch
    Detecting LLM-Generated Text with Classical Machine Learning – Results and Insights

    A scikit-learn TF‑IDF + LinearSVC pipeline can distinguish modern LLM‑generated Chinese text from human‑written text with ~85% sentence‑level accuracy, demonstrating that “classical” ML still works for AI‑text detection.

  • Project
    aa0101181514/tw-legal-rag

    A CLI and MCP server that enables semantic retrieval of over 22 million Taiwan court judgments to provide grounded, verifiable context for AI assistants.