Claude MPM (Multi-Agent Project Manager) is a powerful orchestration framework for Claude Code (CLI) that enables multi-agent workflows, session management, and real-time monitoring through a streamlined Rich-based interface.
You need Claude Code CLI (v1.0.92+), NOT Claude Desktop (the app). All MCP integrations work with Claude Code's CLI interface only.
- Minimum version: v1.0.92 (hooks support)
- Recommended: v2.0.30+ (latest features)
- Install from: https://docs.anthropic.com/en/docs/claude-code
- Claude Code (CLI): Command-line interface for developers with MCP support, hooks, and automation
- Claude Desktop (App): GUI application for general use without CLI features
Claude MPM requires the CLI version for its orchestration capabilities.
# Recommended: Install with monitoring dashboard
pipx install "claude-mpm[monitor]"
# Or with pip
pip install "claude-mpm[monitor]"
# Verify installation
claude-mpm --version
claude-mpm doctorSee Installation Guide for detailed instructions.
- Python: 3.8+ (3.11+ recommended)
- Claude Code CLI: v1.0.92 or higher
- Operating System: macOS, Linux, or Windows
- Disk Space: ~50MB for base installation
- RAM: Minimum 2GB available
pipx is recommended for isolated installations:
# pipx (recommended) - isolated environment
pipx install "claude-mpm[monitor]"
# pip - global/virtualenv installation
pip install "claude-mpm[monitor]"Benefits of pipx:
- ✅ Isolated environments prevent dependency conflicts
- ✅ Automatic PATH configuration
- ✅ Easy upgrades with
pipx upgrade claude-mpm - ✅ Fully supported as of v4.10.0+
Recommended:
[monitor]- Full monitoring dashboard (Socket.IO, web server)kuzu-memory- Advanced memory management (separate package)mcp-vector-search- Semantic code search (separate package)
Most users don't need:
[mcp]- Additional MCP services (mcp-browser, mcp-ticketer)
# Interactive mode (recommended for beginners)
claude-mpm
# With monitoring dashboard
claude-mpm run --monitor
# Resume previous session
claude-mpm run --resume
# Use semantic code search
claude-mpm search "authentication logic"See Getting Started for complete usage.
Agents are specialized AI assistants that handle specific tasks:
- PM (Project Manager): Orchestrates workflow and delegates to specialists
- Engineer: Software development and implementation
- QA: Testing and quality assurance
- Documentation: Creates and maintains docs
- Security: Security analysis and implementation
- Research: Code analysis and research
See Agent System Overview for details.
# Start with monitoring
claude-mpm run --monitor
# Dashboard opens automatically at http://localhost:5000The dashboard shows:
- Live agent activity
- File operations
- Session management
- Performance metrics
See Monitoring Guide for details.
The Resume Log System automatically generates structured 10k-token logs when approaching Claude's context window limits, enabling seamless session continuity.
Key features:
- 🎯 Warnings at 70%, 85%, and 95% thresholds
- 📋 Structured logs for session resumption
- 🔄 Automatic context preservation
- ⚙️ Zero-configuration operation
See Resume Logs Guide for complete documentation.
~/.claude-mpm/ # User-level configuration
├── configuration.yaml # Main configuration
├── agents/ # Custom agents
└── skills/ # Custom skills
.claude-mpm/ # Project-level configuration
├── configuration.yaml # Project config
├── agents/ # Project agents
├── skills/ # Project skills
└── resume-logs/ # Session resume logs
See Configuration Reference for details.
# Interactive configuration
claude-mpm configure
# Manual configuration
edit ~/.claude-mpm/configuration.yamlSee User Guide - Configuration for options.
# Interactive skills management
claude-mpm configure
# Choose option 2: Skills Management
# Skills are stored in:
# - Bundled: ~/.local/share/claude-mpm/skills/
# - User: ~/.config/claude-mpm/skills/
# - Project: .claude-mpm/skills/See Skills Guide for details.
# Run comprehensive diagnostics
claude-mpm doctor
# Check verbose output
claude-mpm doctor --verbose
# Check specific component
claude-mpm doctor --checks installationCommon issues:
- Claude Code not installed: Install from https://docs.anthropic.com/en/docs/claude-code
- Wrong version: Upgrade to v1.0.92+ with
claude upgrade - PATH issues: Ensure
claudecommand is accessible - Python version: Requires Python 3.11+ (kuzu-memory dependency)
See Troubleshooting Guide for solutions.
# Check for Claude MPM updates
claude-mpm doctor --checks updates
# Check Claude Code version
claude --version
# Update Claude MPM
pipx upgrade claude-mpm # if installed with pipx
pip install --upgrade claude-mpm # if installed with pipUpdates are automatically checked on startup (configurable in configuration.yaml).
Large conversation histories can consume 2GB+ of memory. Clean up old conversations:
# Clean up old conversation history
claude-mpm cleanup-memory
# Keep only recent conversations (last 7 days)
claude-mpm cleanup-memory --days 7See User Guide - Memory Management for details.
# Verify all MCP services
claude-mpm verify
# Auto-fix issues
claude-mpm verify --fix
# Verify specific service
claude-mpm verify --service kuzu-memory
# Get JSON output for automation
claude-mpm verify --jsonSee MCP Gateway Documentation for details.
Yes! Claude MPM supports three tiers of agents:
- System: Bundled agents (read-only)
- User: Custom agents in
~/.claude-mpm/agents/ - Project: Project-specific agents in
.claude-mpm/agents/
See Creating Agents for step-by-step guide.
Agents learn project-specific patterns using a simple list format and can update memories via JSON response fields:
remember: Incremental updates (add new learnings)MEMORIES: Complete replacement
Initialize with claude-mpm memory init.
See Memory Integration for technical details.
The MCP Gateway enables integration with external tools and services through the Model Context Protocol:
- Custom tool development
- Protocol-based communication
- Extensible architecture
- Support for third-party MCP servers
See MCP Gateway Documentation for details.
- Fork the repository
- Create a feature branch
- Make changes with tests
- Run quality checks:
make quality - Submit pull request
See Developer Guide and Contributing Guidelines for details.
Yes! kuzu-memory is a recommended partner product that provides:
- 🧠 Persistent project-specific knowledge graphs
- 🎯 Intelligent prompt enrichment
- 📊 Structured storage of project knowledge
- 🔄 Seamless zero-configuration integration
Install with: pipx install kuzu-memory
See README - Recommended Partner Products for details.
Yes! mcp-vector-search is another recommended partner product that enables:
- 🔍 Semantic code search by intent
- 🎯 Context-aware code discovery
- ⚡ Fast indexing for large codebases
- 📊 Pattern recognition and refactoring opportunities
Install with: pipx install mcp-vector-search
Use with: claude-mpm search "authentication logic" or /mpm-search in Claude Code sessions.
Claude MPM includes MCP Gateway which supports:
- Built-in: mcp-browser, mcp-ticketer, filesystem, github
- Partner: kuzu-memory, mcp-vector-search
- Custom: Any MCP-compliant server
See MCP Gateway Documentation for integration details.
Claude MPM v4.8.2+ includes major performance optimizations:
- 91% latency reduction in hook system
- Git branch caching with 5-minute TTL
- Non-blocking HTTP fallback
- 50-80% overall improvement
If still slow:
- Check network connectivity (MCP gateway initialization)
- Disable unused agents in configuration
- Reduce monitoring frequency if using
--monitor
Use the Resume Log System (v4.17.2+):
- Automatic warnings at 70%/85%/95% thresholds
- 10k-token structured logs preserve context
- Seamless session resumption
- Zero configuration required
See Resume Logs Guide for details.
Yes, configure in configuration.yaml:
# Disable auto-save
session:
auto_save:
enabled: false
# Disable resume logs
context_management:
resume_logs:
enabled: false
# Disable monitoring
monitoring:
enabled: falseSee Configuration Reference for all options.
- Getting Started Guide - Installation and first steps
- User Guide - Complete feature documentation
- Troubleshooting Guide - Common issues and solutions
- Developer Guide - Technical documentation
- Agent System - Agent development guide
Still have questions? Check the Documentation Hub or open an issue on GitHub.