AgentStack
MCP unreviewed MIT Self-run

Cas

mcp-codingagentsystem-cas · by codingagentsystem

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

No reviews yet
0 installs
12 views
0.0% view→install

Install

$ agentstack add mcp-codingagentsystem-cas

Open-source listing — not yet scanned by AgentStack. Follow the source repository for install instructions.

Security review

⚠ Flagged

1 finding(s); flagged for manual review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures
  • high Pipes remote content directly into a shell (remote code execution).

What it can access

  • Network access Used
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Cas? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

██████╗ █████╗ ███████╗ ██╔════╝██╔══██╗██╔════╝ ██║ ███████║███████╗ ██║ ██╔══██║╚════██║ ╚██████╗██║ ██║███████║ ╚═════╝╚═╝ ╚═╝╚══════╝

Multi-agent coding factory with persistent memory.

[](LICENSE) [](https://github.com/codingagentsystem/cas/actions) [](https://github.com/codingagentsystem/cas/releases)

[Factory](#factory) · [Context System](#context-system) · [Quick Start](#quick-start) · [Installation](#installation) · [Architecture](#architecture) · [Contributing](CONTRIBUTING.md)


What is CAS?

CAS is a multi-agent coding factory and persistent context system for AI agents. It has two core capabilities:

  1. Factory — A terminal UI that orchestrates multiple Claude Code instances working in parallel on the same codebase, with a supervisor agent coordinating worker agents across isolated git worktrees.
  1. Context System — An MCP server that gives agents persistent memory, task tracking, rules, and skills across sessions, backed by SQLite and full-text search.

Factory

Factory mode turns your terminal into a multi-agent coding operation. A supervisor agent breaks work into tasks while worker agents execute them in parallel — each in its own git worktree to avoid conflicts.

# Launch the factory TUI
cas

# Launch with 3 workers in isolated worktrees
cas -w 3

How it works

┌─────────────────────────────────────────────────────────┐
│  CAS Factory                                            │
├──────────────────────┬──────────────────────────────────┤
│                      │                                  │
│  Supervisor          │  Worker 1        Worker 2        │
│                      │                                  │
│  Plans EPICs,        │  Executes tasks  Executes tasks  │
│  breaks down work,   │  in isolated     in isolated     │
│  assigns tasks,      │  git worktree    git worktree    │
│  reviews & merges    │                                  │
│                      │                                  │
├──────────────────────┴──────────────────────────────────┤
│  Shared: CAS database (memories, tasks, rules, skills)  │
└─────────────────────────────────────────────────────────┘
  • Supervisor plans work, creates tasks, assigns them to workers, reviews completed work, and merges branches
  • Workers each get their own git worktree and branch — no merge conflicts during parallel execution
  • Shared context — all agents read/write the same CAS database for memories, tasks, rules, and coordination messages
  • Built-in terminal multiplexer — side-by-side or tabbed views of all agent sessions, with a custom VT parser (based on Ghostty)

Factory features

| Feature | Description | |---------|-------------| | Worktree isolation | Each worker gets its own git worktree and branch — parallel edits without conflicts | | Task coordination | Supervisor assigns tasks with dependencies; workers claim, execute, and report back | | Live TUI | Side-by-side or tabbed terminal views of all agents, with real-time status bar | | Message passing | Push-based communication between supervisor and workers via prompt queue | | Session management | Attach, detach, list, and kill factory sessions (cas attach, cas list, cas kill) | | Desktop notifications | Optional alerts when tasks complete or workers hit blockers (--notify) | | Session recording | Record terminal sessions for playback (--record) |

When to use Factory

  • Large features — break an epic into subtasks and parallelize across workers
  • Codebase-wide refactors — workers modify different files simultaneously without conflicts
  • Multi-step workflows — tasks with dependencies execute in the right order
  • Code review — supervisor reviews worker output before merging to the main branch

Context System

CAS runs as an MCP server that gives your agent persistent context across sessions — 50+ tools for memory, tasks, rules, skills, and search.

MCP Tools

When your agent has CAS configured, it can:

# Remember something across sessions
mcp__cas__memory action=remember content="This project uses Zod for validation"

# Create and track tasks
mcp__cas__task action=create title="Implement auth" priority=1

# Search past context
mcp__cas__search action=search query="error handling patterns"

# Create a rule that auto-syncs to .claude/rules/
mcp__cas__rule action=create content="Always validate input at API boundaries"

What persists

| Feature | Description | |---------|-------------| | Memory | Learnings, preferences, and observations that survive across sessions | | Tasks | Work items with dependencies, priorities, and structured progress notes | | Rules | Coding conventions that earn trust through use and auto-sync to your editor | | Skills | Reusable agent capabilities with templates and usage tracking | | Search | Fast full-text search (BM25) across all stored context |

Quick Start

# Install
curl -fsSL https://cas.dev/install.sh | sh

# Initialize in your project
cas init

# Launch the factory TUI
cas

Installation

curl (recommended)

curl -fsSL https://cas.dev/install.sh | sh

Homebrew

brew tap codingagentsystem/cas
brew install cas

Build from source

git clone https://github.com/codingagentsystem/cas.git
cd cas
cargo build --release
# Binary at target/release/cas

CLI

cas                   # Launch the factory TUI
cas -w 3              # Launch with 3 workers
cas serve             # Start MCP server for Claude Code
cas init              # Initialize CAS in your project
cas attach            # Attach to a running factory session
cas list              # List running factory sessions
cas kill              # Kill a factory session
cas config list       # View all configuration options
cas doctor            # Run diagnostics
cas update            # Self-update to latest version
cas login             # Log in to CAS Cloud (optional)
cas cloud sync        # Sync data to/from cloud (optional)

Claude Code Integration

Add to your Claude Code MCP config (.claude/settings.json or project .mcp.json):

{
  "mcpServers": {
    "cas": {
      "command": "cas",
      "args": ["serve"]
    }
  }
}

Architecture

Data storage

CAS stores all data locally in your project:

.cas/
├── cas.db          # SQLite — memories, tasks, rules, skills
├── config.yaml     # Project configuration
└── indexes/
    └── tantivy/    # Full-text search index

Storage tiers:

  • Project (.cas/) — project-specific context
  • Global (~/.config/cas/) — cross-project preferences and learnings

Workspace Crates

| Crate | Purpose | |-------|---------| | cas-cli | CLI binary, MCP server, and factory TUI | | cas-factory | Multi-agent session lifecycle and coordination | | cas-factory-protocol | Message protocol between supervisor and workers | | cas-pty | PTY management for agent terminal sessions | | cas-mux | Terminal multiplexer layout and rendering | | cas-core | Core logic, hooks, and integrations | | cas-store | SQLite storage layer | | cas-search | Full-text search (BM25 via Tantivy) | | cas-mcp | MCP protocol handlers | | cas-types | Shared data types | | cas-code | Code analysis (tree-sitter) | | cas-diffs | Diff tracking and formatting | | cas-recording | Terminal session recording and playback | | ghostty_vt | Virtual terminal parser (based on Ghostty) |

Built With

  • Rust for performance and reliability
  • SQLite for local-first storage
  • Tantivy for full-text search (BM25)
  • Ratatui for the factory TUI
  • Ghostty VT for terminal emulation
  • rmcp for MCP protocol support

Configuration

CAS is configured via .cas/config.yaml in your project root. Run cas config list to see all options or cas config describe for details on any setting.

Cloud Sync (optional)

CAS works fully offline. Optionally sync your context across devices:

cas login
cas cloud sync

Cloud sync is not required — all core features work locally with SQLite.

Contributing

CAS is source-available under the MIT license. We welcome bug reports and feature suggestions through Issues and Discussions.

See [CONTRIBUTING.md](CONTRIBUTING.md) for details.

License

[MIT](LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.