MCP Standards

--- skill_name: remember-preferences version: 1.0.0 author: MCP Standards Team description: Automatically remember and recall user preferences using AgentDB personal memory tags: [memory, preferences, learning, agentdb] --- # Remember Preferences Skill **Purpose**: Make Claude remember YOUR preferences automatically across all conversations. ## How It Works This skill enables Claude to: 1. **Auto-detect corrections** - When you say "actually, use X not Y", it's automatically stored 2. **Semantic recall** - Before responding, Claude checks personal memory for relevant preferences 3. **Cross-session learning** - Preferences persist across all conversations ## Usage ### Automatic (Zero Config) Just correct Claude naturally: ``` You: "Run pip install pytest" Claude: *executes* You: "Actually, use uv not pip" Claude: ✓ Remembered: "use uv not pip" (python) Next session: You: "Install pytest" Claude: *uses uv automatically* ``` ### Manual (If Needed) Explicitly tell Claude to remember: ``` Remember: Always run tests before committing code Remember: Use TypeScript for new JavaScript files Remember: Prefer functional components in React ``` ## What Gets Remembered ✅ **Tool preferences**: "use uv not pip", "prefer yarn over npm" ✅ **Workflow patterns**: "run tests before commit", "update docs when adding features" ✅ **Code style**: "use TypeScript", "prefer functional components" ✅ **Project conventions**: "follow PEP 8", "use 2 spaces for indentation" ## Technical Details - **Storage**: AgentDB vector memory (semantic search) - **Speed**: <1ms to recall preferences - **Persistence**: Stored in ~/.mcp-standards/agentdb - **Privacy**: 100% local, no cloud ## Commands ### Check what's remembered ``` What preferences have you remembered about me? ``` ### Search specific category ``` What do you remember about Python package management? What preferences do you have about Git? ``` ### See statistics ``` Show me memory stats ``` ## Integration Points This skill automatically integrates with: - Config file parsing (.editorconfig, pyproject.toml) - Tool execution monitoring (Bash, Read, Write, etc.) - Natural language corrections in conversations ## Best Practices 1. **Be explicit the first time**: "Use uv not pip" is clearer than "don't use pip" 2. **Provide context**: "For Python projects, use uv" helps categorization 3. **Trust the system**: You don't need to repeat yourself - it's remembered after the first correction 4. **Review occasionally**: Ask "what do you remember?" to verify ## Example Workflow ``` Session 1: You: "Run the Python tests" Claude: pytest tests/ You: "Actually, use uv run pytest" Claude: ✓ Remembered Session 2: You: "Run the Python tests" Claude: uv run pytest tests/ [uses remembered preference] Session 10: You: "What Python preferences do you have?" Claude: "I remember you prefer: - Use uv for package management (not pip) - Run tests with uv run pytest - Use Python 3.11+ features" ``` ## Troubleshooting **Preference not being remembered?** - Use clear language: "use X not Y" - Check logs: Look for "🎯 Detected correction" messages **Wrong preference recalled?** - More specific context helps: "for Python projects, use X" - Categories help filtering: python, git, docker, etc. **Want to forget something?** - Currently: Preferences persist forever (future: expiration/override) - Workaround: Correct it again with new preference ## Future Enhancements - Confidence scores (promote after 2-3 occurrences) - Preference conflicts detection - Cross-project learning (global vs project-specific) - Preference expiration (time-based decay) --- **Remember**: The best correction is the one you only have to make once.