High-Taste

An MCP (Model Context Protocol) server that enforces coding style decisions based on high-taste and convention rather than semantic correctness. Following the "Effective C++" tradition, High-Taste provides rules with examples and rationale to maintain consistent, readable code.

Overview

High-Taste helps development teams maintain code quality by:

• Checking new code against established high-taste rules (like a linter for style preferences)
• Learning from expert refactoring by analyzing before/after diffs to create new rules
• Integrating seamlessly with Claude Code and VS Code via MCP

Unlike traditional linters that focus on syntax and semantics, High-Taste enforces subjective but important style decisions that make code more maintainable, readable, and consistent with team preferences.

Core Functions

taste_check

Analyzes Python files against existing high-taste rules and reports violations with specific line numbers and rule references.

Input: List of file contents (Python code) Output: Linting-style errors with rule numbers and descriptions

taste_acquire

Learns new high-taste rules by analyzing diffs where experienced programmers have refactored code originally written by junior developers or AI.

Input: Set of git diffs showing before/after code changes Output: New rules created or confirmation that existing rules already cover the patterns

Implementation Plan

Phase 1: Core Architecture

1. Rule System Design

   • Create rules/ directory with markdown files (one rule per file)
   • Implement rule parser to extract patterns, examples, and metadata
   • Design rule matching engine using AST analysis for semantic patterns
   • Create rule numbering/ID system for references

2. MCP Server Foundation

   • Set up MCP server infrastructure with proper protocol handling
   • Implement high-taste_check function with file content analysis
   • Implement taste_acquire function with diff analysis
   • Create standardized JSON output formats for both functions

Phase 2: Pattern Recognition

3. high-taste_check Implementation

   • Build AST-based pattern matcher for Python code structures
   • Implement rule violation detection with line number reporting
   • Create rule severity system (error, warning, suggestion)
   • Add support for incremental checking (only changed lines)

4. taste_acquire Implementation

   • Develop diff parser to extract before/after code patterns
   • Create pattern generalization algorithm (specific → general rules)
   • Implement rule conflict detection and resolution
   • Build automatic rule description generation

Phase 3: Integration

5. IDE Integration

   • Configure MCP server for Claude Code integration
   • Set up VS Code MCP client configuration
   • Create configuration files for both environments
   • Test end-to-end workflow in both IDEs

6. Rule Management

   • Implement rule CRUD operations (create, read, update, disable)
   • Add rule versioning and change tracking
   • Create rule validation and testing framework
   • Build rule export/import functionality

Phase 4: Advanced Features

7. Intelligence Enhancements

   • Add machine learning for better pattern recognition
   • Implement context-aware rule application
   • Create rule recommendation system
   • Add support for custom rule templates

8. Monitoring and Metrics

   • Track rule violation frequency and trends
   • Measure rule effectiveness and adoption
   • Create dashboard for team rule compliance
   • Add automated rule quality assessment

Installation

Install from PyPI

# Install using pip
pip install high-taste

# Or install using uv
uv pip install high-taste

# Or run directly with uvx (no installation needed)
uvx high-taste --help

Install from source

git clone https://github.com/alexdong/high-taste.git
cd high-taste
uv pip install -e .

Usage

Command Line Interface

High-Taste provides a CLI for standalone usage and testing:

# List all available taste rules
high-taste rules

# Check Python files for taste violations
high-taste check file1.py file2.py

# Start the MCP server
high-taste serve

# Enable debug logging
high-taste --debug check file.py

Using with uvx

You can run High-Taste directly without installation using uvx:

# Run any command with uvx
uvx high-taste rules
uvx high-taste check mycode.py
uvx high-taste serve

MCP Integration

To use High-Taste as an MCP server with Claude Code or VS Code:

1. For Claude Code: Copy configs/claude-code.json to your Claude Code configuration directory
2. For VS Code: Copy configs/vscode.json to your VS Code configuration directory

Then the taste_check and taste_acquire tools will be available in your IDE.

Project Structure

./
├── src/
│   ├── high_taste/         # Main MCP server code
│   │   ├── server.py       # MCP server implementation
│   │   ├── rules/          # Rule management and matching
│   │   ├── parsers/        # AST and diff parsing
│   │   └── utils/          # Helper functions
├── rules/                  # High-Taste rules (markdown files)
│   ├── 001-assertions-over-exceptions.md
│   ├── 002-load-resources-once.md
│   ├── 003-eliminate-deep-nesting.md
│   ├── 004-dynamic-data-discovery.md
│   ├── 005-structured-return-types.md
│   ├── 006-database-transaction-safety.md
│   └── ...
├── tests/                  # Test files
├── examples/               # Example configurations
├── docs/                   # Documentation
└── configs/                # MCP client configurations
    ├── claude-code.json
    └── vscode.json