Status: 0.4.0rc17 (Release Candidate)
CCCC is a local-first multi-agent collaboration kernel that coordinates AI agents like a modern IM.
Key features:
- 🤖 Multi-runtime support — Claude Code, Codex CLI, Droid, OpenCode, Copilot, and more
- 📝 Append-only ledger — Durable history as single source of truth
- 🌐 Web-first console — Mobile-friendly control plane
- 💬 IM-grade messaging — @mentions, reply/quote, read receipts
- 🔧 MCP tool surface — 38+ tools for reliable agent operations
- 🔌 IM Bridge — Telegram, Slack, Discord, Feishu, DingTalk
# Install
pip install --index-url https://pypi.org/simple \
--extra-index-url https://test.pypi.org/simple \
cccc-pair==0.4.0rc17
# Start
ccccOpen http://127.0.0.1:8848/ to access the Web UI.
📚 Read the Docs — Full guides, reference, and API documentation.
Copy this prompt to your AI assistant (Claude, ChatGPT, etc.):
Please help me install and start CCCC (Claude Code Collaboration Context) multi-agent collaboration system.
Steps:
Install cccc-pair:
pip install --index-url https://pypi.org/simple \ --extra-index-url https://test.pypi.org/simple \ cccc-pair==0.4.0rc17After installation, start CCCC:
ccccTell me the access URL (usually http://localhost:8848/ui/)
If you encounter any errors, please help me diagnose and resolve them.
pip install --index-url https://pypi.org/simple \
--extra-index-url https://test.pypi.org/simple \
cccc-pair==0.4.0rc17git clone https://github.com/dweb-channel/cccc
cd cccc
pip install -e .uv venv -p 3.11 .venv
uv pip install -e .
uv run cccc --helpRequirements: Python 3.9+, macOS / Linux / Windows
| Concept | Description |
|---|---|
| Working Group | Collaboration unit with durable history (like a group chat) |
| Actor | An agent session (PTY or headless) |
| Scope | A directory attached to a group |
| Ledger | Append-only event stream |
| CCCC_HOME | Runtime home, default ~/.cccc/ |
CCCC supports multiple agent runtimes:
cccc runtime list --all # List available runtimes
cccc setup --runtime <name> # Configure MCPAuto MCP setup: claude, codex, droid, amp, auggie, neovate, gemini
Manual setup: cursor, kilocode, opencode, copilot, custom
To set up multi-agent collaboration on a project:
# Attach to your project directory
cd /path/to/repo
cccc attach .
# Setup MCP for your runtime
cccc setup --runtime claude
# Add actors (first enabled actor becomes foreman)
cccc actor add foreman --runtime claude
cccc actor add peer-1 --runtime codex
# Start the group
cccc group startThe bundled Web UI provides:
- Multi-group navigation
- Actor management (add/start/stop/restart)
- Chat with @mentions and reply
- Embedded terminal per actor
- Context & automation settings
- IM Bridge configuration
Bridge your working group to IM platforms:
cccc im set telegram --token-env TELEGRAM_BOT_TOKEN
cccc im startSupported: Telegram | Slack | Discord | Feishu/Lark | DingTalk
cccc doctor # Check environment
cccc groups # List groups
cccc use <group_id> # Switch group
cccc send "msg" --to @all
cccc inbox --mark-read
cccc tail -n 50 -f
cccc daemon status|start|stopPlace PROJECT.md at your repo root as the project constitution. Agents read it via cccc_project_info MCP tool.
The Web UI has high privilege. For remote access:
- Set
CCCC_WEB_TOKENenvironment variable - Use an access gateway (Cloudflare Access, Tailscale, WireGuard)
History: v0.3.x → v0.4.x
v0.3.x (tmux-first) proved the concept but hit limits:
- No unified ledger — Messages in multiple files caused latency
- Actor count limit — tmux layout limited to 1–2 actors
- Weak agent control surface — Limited autonomy
- No first-class remote access — Web control plane needed
v0.4.x introduces:
- Unified append-only ledger
- N-actor model
- MCP control plane with 38+ tools
- Web-first console
- IM-grade messaging
Legacy version: cccc-tmux
Apache-2.0