vaddisrinivas/c3
WhatsApp x Claude Code — AI agent framework via MCP. Multiplayer games, meal planning, reminders. Plugins are Markdown files.
Platform-specific configuration:
{
"mcpServers": {
"c3": {
"command": "npx",
"args": [
"-y",
"c3"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
WhatsApp AI agent framework. Define apps in Markdown + JSON. Claude runs them on WhatsApp.
[](https://github.com/vaddisrinivas/c3-py/actions) [](https://www.python.org/downloads/) [](LICENSE)
git clone https://github.com/vaddisrinivas/c3-py && cd c3-py
docker compose up -d && docker attach c3
# Login to Claude (once), scan WhatsApp QR (once). Done.---
---
A framework with three layers:
| Layer | What it does | You touch it? | |---|---|---| | Core | RBAC, session management, MCP (Model Context Protocol) server, bridge lifecycle | No | | Apps | Markdown + JSON, not code — define what your app does and who can use it | Yes — this is where you build | | WhatsApp | Baileys (WhatsApp Web library) bridge, message types, polls, media, reactions | No |
You write apps. The core runs them. WhatsApp is the transport.
---
An app is a directory:
my-app/
├── app.json ← who can use it, what tools it gets
├── CLAUDE.md ← what it does (prompt)
└── skills/*.md ← optional on-demand skills{
"name": "my-app",
"description": "What this app does",
"trust_level": "community",
"sandboxed": true,
"access": {
"commands": { "/start": ["hosts"] },
"dm": ["hosts"],
"group": []
},
"allowed_tools": ["reply", "send_poll", "memory_read", "memory_write"],
"allowed_resources": ["c3://memory/my-app/*"]
}# My App
You do X via WhatsApp.
## RULES — ABSOLUTE
-Loading reviews...