MCP + Rules: Dual-Driven AI for Compliant HarmonyOS App Development

Problem: Syntax Drift in ArkTS Development

General-purpose large language models struggle with HarmonyOS ArkTS development because the ArkTS corpus is limited. The massive amount of TypeScript code in training data causes "syntax drift," where the model produces code that looks like TypeScript but violates ArkTS-specific syntax, APIs, and performance requirements.

Solution: CodeGenie with MCP + Rules

CodeGenie addresses this by combining two mechanisms:

MCP (Model Context Protocol) : Connects the AI to up-to-date knowledge bases — official HarmonyOS documentation, enterprise best practices, design specs (Figma), AST parsers (tree-sitter), and team wikis (Context7).

Rules : Enforces coding standards at three levels (global default, global custom, project-level) with one-click Markdown import, visual creation, and built-in ArkTS lint rules.

Practical Demonstration: Fixing Deprecated APIs

The article walks through a real scenario where the developer instructs CodeGenie:

"Check the file for API deprecation issues and query the knowledge base for fixes."

Step 1: Rule Loading — AI's "Onboarding"

Before generating code, the AI reads the local Rules set, which includes ArkTS-specific lint rules. This ensures all subsequent modifications comply with ArkTS syntax standards.

Step 2: Intelligent Diagnosis — Built-in Check Tool

The AI invokes the built-in check_editor_errors tool, which uses the latest SDK metadata to identify deprecated interfaces in the file.

Step 3: Knowledge Retrieval — Knowledge Base Tool

Upon detecting issues, the AI automatically calls the knowledge base tool to fetch relevant information. The knowledge base contains official API reference documents and official best practices.

Step 4: Compliant Generation — Fusion of Rules, Checks, and Knowledge

The AI combines the Rules set , built-in check tool , and knowledge base tool to automatically summarize an API upgrade plan and generate standards-compliant code.

Core Capabilities: MCP + Rules Dual-Drive

MCP: Ecosystem Integration

MCP acts as a "data cable" that lets the AI read specified knowledge bases in real time. Examples of integrated MCP servers:

HarmonyOS Knowledge Base MCP Server : Indexes official docs and internal best practices.

Figma MCP : Reads design specs and generates UI code.

tree-sitter : Parses AST for cross-file refactoring.

Context7 MCP : Connects team docs, API docs, and issue history.

Rules: Quality Foundation

Rules prevent "wild code" that increases review and refactor costs. Features:

Flexible configuration : One-click Markdown import, visual rule creation, default ArkTS constraints out of the box.

Hierarchical management :

Global default rules: syntax constraints and partial API usage specs.

Global custom rules: team-wide standards.

Project-level rules: project-specific adaptations.

Dynamic invocation : At runtime, the AI selects the appropriate Agent and model, then dynamically calls rule tools based on context, ensuring every generated line adheres to the defined standards.

Summary

On the CodeGenie platform, MCP support breaks the IDE's physical boundaries, turning the AI assistant from a closed black box into an open ecosystem participant. Rules configuration eliminates "wild code," guaranteeing that generated code meets team standards. Together, they provide both an open ecosystem for unlimited context and a solid shield for code quality. Developers can now configure MCP and Rules in CodeGenie to give AI both broad full-stack vision and rigorous compliance awareness, building high-quality, high-performance HarmonyOS applications through configuration-driven development.