Transform Claude Code into a personalized AI assistant that remembers your preferences, automates your workflows, and integrates seamlessly with your development environment.
This is a production-ready persistent memory system for Claude Code that provides:
- Persistent Context: Claude remembers your preferences, project structure, and workflow patterns across all sessions
- Automated Workflows: Pre-configured automation for backups, syncs, and scheduled tasks
- iTerm2 Deep Integration: Expert-level terminal optimization built into every response
- Tool Orchestration: Seamless integration with 15+ MCP servers and development tools
- Smart Session Management: Context loading, project switching, and progress tracking
- Professional Prompt Engineering: Battle-tested prompt templates for research, code, content, and strategy
Without this system, every Claude session starts from scratch. You spend time:
- Re-explaining your preferences and project structure
- Manually tracking tasks and context
- Repeating the same automation patterns
- Looking up commands and file locations
With this system, Claude becomes a true AI partner that:
- ✅ Loads your entire context in < 3 seconds
- ✅ Suggests optimal iTerm2 layouts automatically
- ✅ Remembers your coding style and project patterns
- ✅ Tracks tasks and progress across sessions
- ✅ Executes complex workflows with natural language
- ✅ Backs up your work automatically to S3
- CLAUDE.md: Your AI's long-term memory (preferences, projects, workflows)
- PROJECT-REGISTRY.md: Central index of all active projects
- WORKING-CONTEXT.md: Current session state and progress
- Automated S3 Backups: Daily backups at 2 AM to AWS S3
Pre-configured MCP servers for:
- Perplexity: Web research and current information
- Airtable: Project management and backlog tracking
- Gmail: Email management and automation
- Puppeteer: Web scraping and browser automation
- Memory: Knowledge graphs and entity tracking
- Obsidian: Note-taking and knowledge management
- And 10+ more...
Built-in LaunchAgents for:
- Daily S3 backups (2 AM)
- GitHub ↔ Airtable sync (9 AM)
- Airtable sync (every 15 minutes)
- Daily note creation
- Weekly reviews and cost tracking
Claude automatically suggests:
- Split pane layouts for development
- Triggers for build notifications
- Dynamic profiles for each project
- tmux integration for remote work
- Custom keyboard shortcuts
Pre-built slash commands for:
/research [topic]- Comprehensive research documents/plan-feature [description]- TDD implementation planning/code-review [file]- Thorough code reviews/optimize [file]- Performance analysis/explain-code [path]- Architecture explanations/backlog- Quick backlog capture
Battle-tested 3-phase protocol:
- Core Protocol: 8-component foundation (Objective, Context, Persona, Examples, Deliverables, Constraints, Reasoning, Validation)
- Modular Augments: Task-specific modules (RAG, Decomposition, Style Guide, Framework)
- Execution Protocols: Static (simple) vs Dynamic (complex with rejection loops)
- macOS (tested on Sonoma/Sequoia)
- Claude Code installed
- iTerm2 (recommended, but Terminal.app works)
- AWS CLI (for S3 backups, optional)
- Homebrew (for dependencies)
# 1. Clone this repository
git clone https://github.com/mikefinneran-sudo/claude-md-setup.git
cd claude-md-setup
# 2. Run the installer
./install.sh
# 3. Follow the interactive prompts to configure:
# - Your name and email
# - Project directories
# - S3 backup settings (optional)
# - MCP servers to enable
# - Automation preferences
# 4. Start a new Claude session
claude
# 5. Verify setup
claude> Load context: What am I working on?That's it! Claude now has persistent memory.
Start a new session with full context:
claude
> Load context: What am I working on?Switch between projects:
> Continue WalterSignal - show status and next actionsSave progress:
> Update working context with today's progressBacklog capture:
> Backlog: Add feature for user authentication with OAuthResearch:
> Research the latest Next.js 14 server components best practicesiTerm2 optimization (automatic):
> I need to run tests
# Claude automatically suggests: "Let me set up a split pane - code on left, test watcher on right..."> /plan-feature User profile with avatar upload
> /code-review src/components/UserProfile.tsx
> /optimize src/utils/dataProcessor.ts
> /research GraphQL vs REST for mobile apps
> /backlogType these anywhere:
;ctx→ Loads session context;save→ Saves progress;weekly→ Weekly review;backlog→ Add to backlog
~/.claude/
├── CLAUDE.md # Main memory file (auto-loads every session)
├── WORKING-CONTEXT.md # Current session state
├── PROJECT-REGISTRY.md # Project index
├── AUTOMATION_LOCATIONS.md # All scheduled tasks
├── commands/ # Slash commands
│ ├── research.md
│ ├── plan-feature.md
│ ├── code-review.md
│ └── ...
├── scripts/ # Automation scripts
│ ├── backup-to-s3.sh
│ ├── github-airtable-sync.py
│ ├── start-session.sh
│ └── ...
├── logs/ # Automation logs
└── templates/ # Document templates
~/Documents/ObsidianVault/ # Primary work hub
├── Projects/ # All active projects
├── Daily Notes/ # Daily work logs
├── Research/ # Research documents
└── Archive/ # Completed projects
~/Library/LaunchAgents/ # Scheduled automations
├── com.claude.s3-backup.plist
├── com.claude.github-sync.plist
└── com.claude.airtable-sync.plist
CLAUDE.md - Main memory (auto-loads)
- User profile and preferences
- Current week focus
- Project locations
- Tool integrations
- Custom commands
- Coding style preferences
PROJECT-REGISTRY.md - Project index
- All active projects
- Status and progress
- Key files and commands
- Team members (if applicable)
WORKING-CONTEXT.md - Session state
- Current task
- Recent progress
- Open questions/blockers
- Next actions
Edit ~/.claude/CLAUDE.md to customize:
## User Profile
- Name: [Your Name]
- Primary Use Case: [Your primary use case]
- Primary Email: [your.email@example.com]
## Current Week Focus
**Week of**: 2025-11-XX
**Primary Project**: [Current project]
**Status**: [Current status]
## Default File Locations
- Primary Storage: [Your main directory]
- Projects: [Projects directory]
- Backups: [Backup location]Automatically backs up:
- All
.claude/files ObsidianVault/(configurable)- Automation logs
- Project files (selective)
Setup:
# Configure AWS credentials
aws configure
# Set S3 bucket
export S3_BACKUP_BUCKET=your-bucket-name
# Test backup
~/.claude/scripts/backup-to-s3.sh
# Enable daily automation (2 AM)
launchctl load ~/Library/LaunchAgents/com.claude.s3-backup.plistRestore:
restore-s3
# Select backup date from interactive menuEnable additional MCP servers:
# Add Perplexity for research
claude mcp add perplexity-research
# Add Puppeteer for web scraping
claude mcp add puppeteer
# Add Airtable for project management
claude mcp add airtable
# List all available servers
claude mcp listSetup complete iTerm2 integration:
~/.claude/scripts/iterm-complete-setup.shThis configures:
- Dynamic profiles for each project
- Custom triggers for notifications
- Status bar with git/npm/python info
- Smart keyboard shortcuts
- Shell integration
Documentation: ~/.config/iterm2/EXPERT_GUIDE.md
Automatically sync your GitHub repos to Airtable for unified project management:
# Manual sync
github-sync
# Check status
github-sync-status
# View logs
github-sync-logs
# Runs automatically daily at 9 AM# Verify CLAUDE.md exists
cat ~/.claude/CLAUDE.md | wc -l
# Check file size (should be 500-700 lines)
du -h ~/.claude/CLAUDE.md
# Restore from backup if needed
restore-s3# List all LaunchAgents
launchctl list | grep -E "claude|com\."
# Check specific automation
launchctl list | grep s3-backup
# View logs
tail ~/.claude/logs/s3-backup.log# Check CLAUDE.md size
wc -l ~/.claude/CLAUDE.md
# Archive old content if > 700 lines
mkdir -p ~/Documents/ObsidianVault/Archive/$(date +%Y-%m)
# Move old sections to archive
# Verify performance
time cat ~/.claude/CLAUDE.md > /dev/null# List all MCP servers
claude mcp list
# Remove and re-add problematic server
claude mcp remove [server-name]
claude mcp add [server-name]
# Check logs
tail ~/.claude/logs/mcp-*.log- Load time: < 3 seconds for full context
- Memory usage: ~5000 tokens (2.5% of 200K window)
- Session startup: < 1 second
- Backup size: ~5-50 MB (configurable)
- Start every session with context load: "Load context: What am I working on?"
- Update working context regularly: Save progress after major milestones
- Weekly reviews: Update current week focus every Monday
- Clean up: Archive completed projects monthly
- One source of truth: ObsidianVault for all work
- Clear naming: Use date prefixes (YYYY-MM-DD - Description - vN)
- Version control: Git commit regularly
- Backups: Verify S3 backups weekly
- MCP first: Use MCP tools before bash commands when available
- Parallel operations: Run independent tasks in parallel
- Context-aware: Reference files with
@filenamefor clarity - Screenshots: Use visual feedback for UI work (Cmd+Ctrl+Shift+4)
- Never commit secrets: Use 1Password, not CLAUDE.md
- Sanitize sharing: Remove emails, API keys, personal data
- Rotate credentials: Quarterly rotation for all API keys
- Backup encryption: Use AWS KMS for S3 backups (optional)
- Review current week focus
- Update PROJECT-REGISTRY.md
- Archive completed projects
- Check automation logs
- Clean up old backups (S3)
- Review and archive old projects
- Update CLAUDE.md (remove stale info)
- Check for MCP server updates
- Rotate API credentials
- Review automation effectiveness
- Update documentation
- Backup to external drive (redundancy)
Contributions welcome! Please:
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Test your changes thoroughly
- Update documentation
- Submit a pull request
Areas for contribution:
- Additional MCP server integrations
- Windows/Linux support
- VSCode integration
- Automation improvements
- Documentation enhancements
Q: Does this work with Windows/Linux? A: Core functionality yes, but automation scripts are macOS-specific (LaunchAgents). PRs welcome for cross-platform support!
Q: How much does this cost? A: The system is free. Optional costs: AWS S3 (~$1-5/month), MCP servers (Perplexity Pro: $20/month), Claude Pro (if using web version).
Q: Is my data private? A: Yes. Everything runs locally. MCP servers only access what you explicitly allow. S3 backups are in your private bucket.
Q: Can I use this with other AI assistants? A: The CLAUDE.md structure works with any AI that supports persistent context files. Automation is Claude Code-specific.
Q: How do I update?
A: git pull origin main && ./install.sh (preserves your customizations)
Q: Can I share my setup publicly?
A: Yes! Use ./scripts/sanitize-for-sharing.sh to remove sensitive data first.
v1.0.0 (2025-11-XX)
- Initial public release
- Complete persistent memory system
- S3 backup automation
- MCP server integrations
- iTerm2 expert integration
- Prompt engineering framework
- Comprehensive documentation
MIT License - See LICENSE file for details
- Anthropic for Claude Code
- iTerm2 team for the amazing terminal
- MCP community for server integrations
- Everyone who contributed ideas and feedback
- Documentation: /docs
- Issues: GitHub Issues
- Discussions: GitHub Discussions
Made with ❤️ for the Claude Code community
Stop starting from scratch. Let Claude remember.