claude-mpm and cas

Both tools offer multi-agent orchestration for Claude with a focus on code, but **bobmatnyc/claude-mpm** provides a broader, SDK-driven framework with a plugin system and multi-channel orchestration, while **codingagentsystem/cas** focuses on persistent memory, tasks, rules, and skills for more coordinated agent interactions; thus, they are largely competitors, though a user might potentially use CAS as a more specialized "plugin" or agent implementation *within* the broader MPM framework.

claude-mpm
74
Verified
cas
45
Emerging
Maintenance 13/25
Adoption 19/25
Maturity 24/25
Community 18/25
Maintenance 13/25
Adoption 8/25
Maturity 13/25
Community 11/25
Stars: 95
Forks: 18
Downloads: 13,499
Commits (30d): 0
Language: Python
License:
Stars: 66
Forks: 7
Downloads:
Commits (30d): 0
Language: Rust
License: MIT
No risk flags
No Package No Dependents

About claude-mpm

bobmatnyc/claude-mpm

Claude Multi-Agent Project Manager — multi-channel orchestration, GitHub-first SDK mode, and plugin system for Claude

Provides 47+ specialized agents, 56 bundled skills, and a hooks system that integrates with Claude Code CLI via plugin marketplace, while leveraging MCP for Google Workspace, Notion, Confluence, Slack, and semantic code search. Implements progressive skill disclosure to optimize token usage and features session resumption with automatic context summarization at configurable thresholds. Built as a service-oriented architecture with 100+ CLI commands, OAuth 2.0 authentication, and encrypted token storage for enterprise workflows.

About cas

codingagentsystem/cas

Multi-agent orchestration for Claude Code. Persistent memory, tasks, rules, and skills that make AI agents actually coordinate.

Implements a supervisor-worker architecture where a coordinator agent plans work and assigns tasks to parallel worker agents, each operating in isolated git worktrees to prevent conflicts during concurrent edits. The Context System runs as an MCP server providing 50+ tools for persistent SQLite-backed memory, task tracking, rules, and full-text search across Claude Code sessions. Built entirely in Rust with a custom VT parser for real-time TUI multiplexing of all agent terminals.

Scores updated daily from GitHub, PyPI, and npm data. How scores work