MCP Standards

# Changelog All notable changes to MCP Standards will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). --- ## 📦 ARCHIVED - [0.2.0-alpha.1] - 2024-11-30 ### Archive Notice **This project has been archived and is provided for reference only.** See [ARCHIVE.md](ARCHIVE.md) for complete details on: - Final project status and achievements - Security audit results (no secrets found) - Complete documentation index - Lessons learned and technical insights ### Added (v0.2.0-alpha.1 - Final Release) - **AgentDB Integration**: Ultra-fast vector memory (<1ms search, 150x faster than SQLite) - **Simple MCP Server**: Reduced from 6,000+ LOC to ~950 LOC (84% reduction) - **Auto-Detection**: Automatic pattern recognition for "use X not Y" preferences - **Zero-Config Design**: 5-minute setup with automatic configuration - **4 Core MCP Tools**: `remember`, `recall`, `list_categories`, `memory_stats` - **Comprehensive Testing**: 5/5 validation tests passing - **Semantic Search**: Context-aware preference matching with vector embeddings - **25+ Claude Skills**: Integration with Claude Code ecosystem ### Changed - **Architecture**: Simplified to focus on core auto-learning functionality - **Storage**: Migrated from SQLite to AgentDB vector memory - **Performance**: <1ms search latency (vs 50ms+ in v1) - **UX**: Zero manual steps (vs 4-5 manual MCP calls in v1) - **Code Size**: 84% reduction in codebase complexity ### Security - ✅ **Security Audit Complete**: No secrets, tokens, or credentials in codebase - ✅ **Privacy Verified**: 100% local storage, no cloud dependencies - ✅ **Git History Clean**: No leaked secrets in commit history - ✅ **Configuration Safe**: `.env.example` is template only --- ## [0.1.0-alpha.1] - 2025-10-15 ### Added - **Self-Learning System**: Automatically learns from corrections and updates CLAUDE.md - **Episode Storage**: Manual knowledge storage with full-text search - **Learned Preferences**: Automatic pattern detection from repeated corrections - **AI Standards Generation**: Auto-generate CLAUDE.md from project config files - **Cost Optimization Integration**: Optional agentic-flow integration for 99.5% cost savings - **Security Features**: Path validation, audit logging, CLAUDE.md-only updates - **Pattern Learning**: Confidence scoring system (0.3, 0.7, 0.9 thresholds) - **9 MCP Tools**: add_episode, search_episodes, list_recent, generate_ai_standards, get_learned_preferences, suggest_claudemd_update, update_claudemd, log_tool_execution, export_to_markdown - **Comprehensive Documentation**: Quick Start Guide, Security Guide, Architecture docs ### Changed - Rebranded from `research-mcp/claude-memory` to `mcp-standards` - Database path: `.claude-memory` → `.mcp-standards` - Server name: `claude-memory` → `mcp-standards` - Repository: `research-mcp` → `airmcp-com/mcp-standards` ### Fixed - MCP server startup with `--directory` flag for proper module resolution - Agentic-flow configuration with correct repository and tracking variables - Claude Desktop config path documentation (macOS) - Installation instructions with proper `uv` commands ### Documentation - Added comprehensive Quick Start Guide with two-system explanation - Added 60-second quickstart tutorial - Clear distinction between Episodes (manual) vs Learned Preferences (automatic) - Practical example: Teaching Claude to use `uv` instead of `pip` - Installation instructions for packaged versions ### Technical - Python 3.10+ with uv package manager - SQLite with FTS5 full-text search - MCP 1.0 protocol compliant - 12/12 smoke tests passing - Local-first architecture (100% privacy) ## [Unreleased] ### Planned - Additional test coverage (integration and unit tests) - Performance benchmarking - Multi-language support for standards generation - Plugin system for custom extractors - Web UI for knowledge base management [0.1.0-alpha.1]: https://github.com/airmcp-com/mcp-standards/releases/tag/v0.1.0-alpha.1