Documentation
Give an agent the keys to ManyChat.
Same story as the landing, written so you can actually wire it. Deep self-host, HTTP transport, and tool maps live in the repository.
What this is
An open-source CLI and Model Context Protocol server. Your agent reads leads, qualifies, tags, updates fields, triggers flows, and follows up. Every send is checked against Meta messaging rules first.
- Messenger
- Telegram
- SMS
Connect
Do not paste JSON by hand unless you want to. Open the install prompts, pick your agent, copy, drop it on the agent. It installs mcp-manychat over stdio and asks for the key only when it writes the config file.
Claude Code
`.mcp.json` in this repo, or `claude mcp add`
Cursor
`~/.cursor/mcp.json` (global) or `.cursor/mcp.json` (this project)
Codex
`~/.codex/config.toml`
OpenCode
`opencode.json` — `mcp.servers` (V2) or `mcp` (V1)
Claude Desktop
`claude_desktop_config.json`
Hermes
`~/.hermes/config.yaml` plus `~/.hermes/.env`
API key
ManyChat → Settings → API → Generate your API Key. Pro account. Shown once. That key is full access to the page. The prompt tells the agent to put it in the client config env (or ~/.hermes/.env for Hermes) and never echo it in the chat.
Safety
validate_message runs before send_text_message and send_content. Outside the 24-hour window, a promo without a tag, or a missing opt-in: refused up front. Borderline cases stop and get tagged for a human.
License
AGPL-3.0-or-later. You can run it commercially, fork it, and self-host it complete. If you run a modified version as a network service, you offer those users the source of that version. Connecting ChatGPT, Claude, or Cursor as a client does not copyleft your chats or your ManyChat account. The hosted dashboard is a separate private codebase; this runtime never requires it.
Source
Canonical tree: github.com/WIZNEOAI/Manychat-MCP. Agents can read /llms.txt.
Not affiliated with, endorsed by, or sponsored by ManyChat, Inc.