MCP Standards

# Quick Start: Simple Personal Memory MCP **Goal**: Make Claude remember YOUR preferences automatically. Zero config, zero manual steps. ## Installation (5 Minutes) ### Step 1: Setup AgentDB ```bash cd /path/to/mcp-standards # Install dependencies npm install # Setup AgentDB npm run setup ``` This will: - ✓ Install AgentDB - ✓ Create ~/.mcp-standards/agentdb directory - ✓ Test AgentDB functionality ### Step 2: Configure Claude Desktop Add to `~/Library/Application Support/Claude/claude_desktop_config.json`: ```json { "mcpServers": { "mcp-standards-simple": { "command": "uv", "args": [ "run", "--directory", "/ABSOLUTE/PATH/TO/mcp-standards", "python", "-m", "mcp_standards.server_simple" ] } } } ``` **Replace `/ABSOLUTE/PATH/TO/` with your actual path!** ### Step 3: Restart Claude Desktop ```bash # Quit Claude Desktop completely # Restart it ``` ### Step 4: Test It! Open Claude Desktop and try: ``` You: Remember: use uv not pip for Python projects Claude: ✓ Remembered: 'use uv not pip for Python projects' (python) You: What do you remember about Python? Claude: I remember you prefer: - Use uv not pip for Python projects ``` **That's it! You're done!** --- ## Daily Usage ### Automatic Learning (Recommended) Just correct Claude naturally - it learns automatically: ``` You: "Install pytest" Claude: pip install pytest You: "Actually, use uv not pip" Claude: ✓ Remembered Next time: You: "Install pytest" Claude: uv pip install pytest [uses remembered preference] ``` ### Manual Memory (Optional) You can also explicitly tell Claude to remember: ``` Remember: Always run tests before committing Remember: Use TypeScript for new JavaScript files Remember: Prefer functional components in React ``` ### Query Memory ``` What preferences have you remembered? What do you remember about Git? Show me memory stats List all categories ``` --- ## How It Works ### Architecture ``` User corrects Claude ↓ Auto-detection hook triggers ↓ Stores in AgentDB (semantic vector memory) ↓ Next session: Claude queries AgentDB before responding ↓ Uses remembered preference automatically ``` ### What Gets Remembered ✅ **Tool preferences**: "use uv not pip", "prefer yarn over npm" ✅ **Workflow patterns**: "run tests before commit" ✅ **Code style**: "use TypeScript", "prefer functional components" ✅ **Project conventions**: "follow PEP 8" ### Categories (Auto-Detected) - `python` - Python-related preferences - `javascript` - JavaScript/Node.js preferences - `git` - Git workflow preferences - `docker` - Docker/container preferences - `testing` - Testing preferences - `general` - Everything else --- ## MCP Tools Available ### Personal Memory (NEW) ```javascript // Store preference remember({ content: "use uv not pip", category: "python" }) // Search preferences recall({ query: "package manager", category: "python", top_k: 5 }) // List all categories list_categories() // Get statistics memory_stats() ``` ### Config Standards (EXISTING) ```javascript // Generate minimal CLAUDE.md from project config generate_ai_standards({ project_path: ".", formats: ["claude", "copilot", "cursor"] }) ``` --- ## Troubleshooting ### AgentDB not found ```bash # Install manually npm install -g agentdb # Verify npx agentdb --version ``` ### Preferences not being remembered Check logs in Claude Desktop: - Look for "🎯 Detected correction" messages - Look for "✓ Remembered" confirmations ### AgentDB slow on first startup First startup loads the embedding model (~23MB). Subsequent startups are <10ms. ### Want to reset memory ```bash # Delete AgentDB data rm -rf ~/.mcp-standards/agentdb/* # Restart Claude Desktop ``` --- ## Performance - **Startup**: <10ms (after first model load) - **Search**: <1ms (HNSW vector index) - **Storage**: 100% local (no cloud) - **Memory**: ~50MB (embedding model) --- ## What's Different from v1? | Feature | v1 (Old) | v2 Simple (New) | |---------|----------|-----------------| | **Memory storage** | SQLite keyword search | AgentDB vector search | | **Detection** | Manual MCP calls | Automatic hooks | | **Search speed** | 50ms+ | <1ms | | **Semantic matching** | ❌ No | ✅ Yes | | **Setup** | Complex | 5 minutes | | **Workflow** | 4-5 manual steps | Zero steps | --- ## Next Steps ### Week 1: Use It Yourself Just use Claude normally. Correct it when needed. Watch it learn. ### Week 2: Refinement ``` What preferences have you remembered? ``` Review and see what it learned. The system improves as you use it. ### Future Features (Coming) - ✨ Confidence scoring (promote after 2-3 corrections) - ✨ Cross-project learning (global vs local) - ✨ Preference conflicts detection - ✨ Web dashboard for memory management --- ## Questions? - **Documentation**: See `docs/SIMPLE_V2_PLAN.md` - **Skill Guide**: See `.claude/skills/remember-preferences.md` - **Architecture**: AgentDB + Simple hooks + Minimal CLAUDE.md - **Inspiration**: Gunnar's simple side projects + context engineering --- **Remember**: The best preference is the one you only have to correct once. 🚀 **Ready to try it? Run `npm run setup` now!**