CC Swarm

Connect multiple Claude Code instances into a coordinated production swarm.

Turn 3-10 machines running Claude Code into a unified AI workforce. Shared state, task queues, real-time messaging, file coordination — all through native CC tools.

---

The Problem

Claude Code is powerful on one machine. But production work happens across many:

• A render farm for video/audio generation
• A server for backend services and deployments
• A posting machine for social media automation
• A dev box for coding and testing

Each CC works in isolation. No shared context. No task coordination. No way to say "CC1, render this — CC3, post it when done."

The Solution

CC Swarm adds 13 native tools to every Claude Code instance:

You: "Check the swarm"
CC:  ⚡ 3 nodes online — CC1 (engine), CC2 (render), CC3 (posting)
     📋 2 tasks pending, 1 in progress
     💬 5 unread messages

You: "Assign the video render to CC2 and tell CC3 to post it when done"
CC:  ✅ Task assigned to CC2 — Mac Studio
     💬 Message sent to CC3 — "Post video when CC2 completes render"

Quick Start

1. Start the Hub (any machine)

npx @atlas/cc-swarm-hub
# ⚡ CC Swarm Hub running on port 7777
# Dashboard: http://localhost:7777

2. Connect Each CC

Add to ~/.claude/settings.json on each machine:

{
  "mcpServers": {
    "cc-swarm": {
      "command": "npx",
      "args": ["@atlas/cc-swarm-mcp"],
      "env": {
        "SWARM_HUB": "http://192.168.1.100:7777",
        "SWARM_NODE_ID": "my-machine",
        "SWARM_ROLE": "render-farm"
      }
    }
  }
}

3. Done

Every CC now has swarm tools. Ask it to check status, assign tasks, message other nodes, share files, or coordinate workflows.

Tools

Tool	Description
swarm_status	Dashboard view — nodes, tasks, messages
swarm_nodes	List all connected CCs with roles
swarm_assign_task	Create and assign work to a specific CC
swarm_claim_task	Pick up the next available task
swarm_complete_task	Mark a task done with results
swarm_get_tasks	View task queue (filter by status/assignee)
swarm_broadcast	Message all CCs at once
swarm_message	Direct message a specific CC
swarm_get_messages	Read messages sent to this CC
swarm_set_state	Set shared key-value state
swarm_get_state	Read shared state
swarm_share_file	Announce a file is ready for other CCs
swarm_get_files	List all shared files

Dashboard

The hub includes a live web dashboard at http://hub-ip:7777:

• Real-time node status (online/offline)
• Task queue with priorities
• Live WebSocket event feed
• Message history

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   CC1        │     │   CC2        │     │   CC3        │
│   Engine     │     │   Render     │     │   Posting    │
│   + MCP      │     │   + MCP      │     │   + MCP      │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │     HTTP/WS        │                    │
       └────────────────────┼────────────────────┘
                            │
                    ┌───────┴───────┐
                    │  CC Swarm Hub  │
                    │  Port 7777     │
                    │  SQLite + WS   │
                    └───────────────┘

• Hub: Express + WebSocket + SQLite. Zero-config, portable, single process.
• MCP Server: Runs alongside each CC. Translates swarm tools → Hub API calls.
• Transport: HTTP REST for operations, WebSocket for real-time events.

Use Cases

Production Studio

• CC1 (Mac Mini): Runs PM2 services, handles deployments
• CC2 (Mac Studio): Renders video, generates audio, processes images
• CC3 (Mac Mini): Posts to social media, manages distribution

Development Team

• CC1 (Dev Server): Runs tests, CI/CD
• CC2 (Local): Writes code, reviews PRs
• CC3 (Staging): Deploys and monitors

Content Factory

• CC1: Writes content, generates scripts
• CC2: Creates visuals, renders video
• CC3: Posts to all platforms, tracks engagement

What We Built With It

In 3 days with a 3-machine CC Swarm, one person produced:

• 20 audiobooks (200+ hours of content)
• 40 AI-generated music tracks
• 20 video spots for social media
• Multi-platform social campaign (TikTok, YouTube, X)
• Voice cloning + custom VO generation
• 9,000+ videos downloaded and organized
• 8,600+ photos exported and cataloged

That's what happens when AI agents work together.

License

MIT

Built by

Backbone Logic — AI-powered production infrastructure.