A linguistic linter for Traditional Chinese (zh-TW), zhtw-mcp addresses the critical issue of AI models generating text that drifts towards Mainland Chinese (zh-CN) conventions. It enforces Taiwan Ministry of Education (MoE) standards for vocabulary, punctuation, and character shapes. By integrating with AI coding assistants via the Model Context Protocol (MCP), it catches and corrects regional linguistic drift before it reaches users, ensuring adherence to Taiwanese norms.

How It Works

The linter enforces three core Taiwanese standards: the Revised Handbook of Punctuation, Standard Form of National Characters, and cross-strait vocabulary normalization derived from OpenCC datasets. It compiles over 1100 vocabulary rules and 15 casing rules into a binary. Novelty lies in its MCP integration, allowing AI assistants to query for context on ambiguous terms without requiring external API keys, directly addressing AI model biases towards zh-CN.

Quick Start & Requirements

Pre-built binaries are available via shell scripts for macOS/Linux and PowerShell for Windows. Building from source requires Rust 1.91+. The make install command builds, installs to ~/.local/bin, and registers the MCP server. Manual MCP setup is also supported. Official documentation is available for CLI and MCP usage.

Highlighted Details

• Profiles: base (vocabulary, punctuation, casing, grammar, political terms) and strict (full MoE character variants and punctuation).
• Flags: relaxed (lenient punctuation/grammar for UI) and detect_ai (AI writing artifact detection).
• MCP Integration: Enables natural language commands for linting, auto-fixing, quality gating, and AI writing review directly within AI coding assistants.
• Rule Set: Enforces over 1100 vocabulary rules, 15 casing rules, and MoE standards for punctuation and character shapes.

Maintenance & Community

The provided README does not detail specific maintenance contributors, community channels (e.g., Discord, Slack), or roadmap links.

Licensing & Compatibility

Licensed under a permissive MIT-style license, allowing for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The README does not explicitly list known limitations, bugs, or indicate an alpha/beta status, presenting the tool as a production-ready solution. MCP integration requires compatible AI coding assistant environments.