Version 4.17.2 | Claude Multi-Agent Project Manager
Welcome to Claude MPM - a powerful orchestration framework that extends Claude Code with multi-agent workflows, persistent memory, and real-time monitoring.
# 1. Install
pipx install "claude-mpm[monitor]"
# 2. Auto-configure your project
claude-mpm auto-configure
# 3. Run
claude-mpm run --monitorThat's it! Auto-configuration detects your project's stack and deploys the right agents automatically.
- Installation - Install Claude MPM and Claude Code CLI
- Quick Start - Get running in 5 minutes
- Auto-Configuration - Automatic project setup and agent deployment
- User Guide - Features, workflows, and best practices
- Resume Log System - Proactive context management guide
- Troubleshooting - Common issues and solutions
- FAQ - Frequently asked questions
- Single-Tier Agent System - Git-based agent management (NEW in v5.0)
- Monitoring - Real-time agent monitoring
- Ticketing Workflows - Project management integration
- Configuration Reference - Complete configuration options
- Deployment Overview - Deployment strategies and operations
- Architecture Overview - High-level system design
- Single-Tier Design - Agent system architecture (NEW in v5.0)
- Design Decisions - Feature designs and rationale
- Agent Overview - Multi-agent orchestration concepts
- PM Workflow - Project Manager agent patterns
- Agent Patterns - Creating effective agents
- Creating Agents - Step-by-step agent development
- API Reference - Complete API documentation
- CLI Reference - Command-line interface
- Agent Sources API - Technical API reference (NEW in v5.0)
- Developer Architecture - System design and core concepts
- Extending - Build custom agents, hooks, and services
- Contributing - Development setup and contribution guidelines
- Resume Log Examples - Real-world resume log workflows and tutorials
Auto-Configuration: Automatic stack detection and agent deployment (since v4.10.0)
- Detects Python, Node.js, Rust, Go, and more
- Identifies frameworks (FastAPI, Next.js, React, etc.)
- Recommends agents with confidence scores (80%+ default)
Multi-Agent System: 15+ specialized agents for comprehensive project management
- Project Manager (PM) orchestrates workflow
- Research, Engineer, QA, Documentation specialists
- Language-specific coding agents (Python, TypeScript, Go, Rust, etc.)
Agent Memory: Persistent learning with JSON response fields
- Project-specific knowledge graphs
- Cross-session learning and context
- Automatic context enrichment
Local Process Management: Professional-grade local deployment (since v4.13.0)
- Three-tier health checks (HTTP, process, resource)
- Auto-restart with exponential backoff
- Memory leak detection and circuit breaker
- Configuration-driven via YAML
Resume Log System: Proactive context management (since v4.17.2)
- Graduated thresholds at 70%/85%/95% (60k token buffer)
- Automatic 10k-token structured logs for session continuity
- Seamless resumption with full context preservation
- Zero-configuration automatic operation
Session Management: Automatic saving and manual pause/resume with full context
- NEW: Auto-save every 5 minutes (configurable 60-1800s)
- Graceful shutdown with final save - no data loss
- Manual pause:
claude-mpm mpm-init pause - Resume:
claude-mpm mpm-init resume - Automatic change detection
Real-Time Monitoring: Live dashboard shows agent collaboration
- WebSocket-based updates
- Agent activity tracking
- Performance metrics
Performance: 91% latency reduction in hook system (since v4.8.2)
- Git branch caching with 5-minute TTL
- Non-blocking HTTP fallback
- 50-80% overall improvement
Security: Comprehensive input validation and filesystem restrictions
Service-Oriented: Five specialized service domains with clear contracts
- Quick Reference: Use the documentation links above
- Troubleshooting: Check user/troubleshooting.md
- GitHub Issues: Report bugs or request features
- API Docs: See developer/api-reference.md
Quick Tip: Start with Getting Started if you're new, or jump to Architecture for system internals.