ran-agent：邮件驱动的多功能 AI 智能体

ran-agent 是一个以邮件为指令入口、可配置智能体群组的多功能 AI 智能体系统。

背景

OpenClaw 火了之后，我想真正搞懂这类 AI 编程助手产品背后的工作原理——Skill、Tool Call、Agent Loop 是怎么跑起来的。于是我借助 Claude Code 从零开发了这个项目。

由于在 Windows 环境下开发，无法使用 macOS 特有的系统级多端调用能力，我换了一个思路：用邮件作为指令通道——用户向指定邮箱发送命令，智能体接收、解析并执行，再将结果回复到邮箱。

主要功能

• 邮件指令触发 — 向指定邮箱发送自然语言命令，触发智能体执行任务
• 智能体群组配置 — 在管理页面配置多个 Agent，协作完成复杂任务
• 工具调用支持 — 集成 MCP 等工具调用协议，扩展 Agent 能力边界
• 跨平台运行 — 基于 Python 开发，Windows / macOS / Linux 均可部署

工作原理

用户发送邮件命令
      ↓
邮件监听服务解析指令
      ↓
调度到对应 Agent（或 Agent 群组）
      ↓
Agent 执行任务（含工具调用）
      ↓
结果回复到邮箱

技术选型

模块	技术
Agent 推理	Claude API
工具调用	MCP（Model Context Protocol）
指令通道	SMTP / IMAP（邮件收发）
开发框架	Python

项目地址

github.com/ramadanma/ran-agent

ran-agent is a multi-function AI agent system that accepts commands via email and supports configurable agent groups.

Background

After OpenClaw gained traction, I wanted to understand how AI coding-assistant products actually work under the hood — how Skills, Tool Calls, and Agent Loops are executed. So I built this project from scratch using Claude Code.

Developing on Windows, I didn’t have access to the system-level multi-platform calling capabilities available on macOS, so I took a different approach: using email as the command channel. Users send a natural-language command to a designated mailbox; the agent receives, parses, and executes it, then replies with the result.

Key Features

• Email Command Trigger — Send natural-language commands to a mailbox; the agent picks them up and runs
• Agent Group Configuration — Set up multiple agents on a management page and have them collaborate on complex tasks
• Tool Call Support — Integrates MCP and other tool-call protocols to extend agent capabilities
• Cross-platform — Python-based; runs on Windows, macOS, and Linux

How It Works

User sends an email command
          ↓
Mail listener parses the instruction
          ↓
Dispatched to the relevant Agent (or Agent group)
          ↓
Agent executes the task (with tool calls as needed)
          ↓
Result is sent back by email

Tech Stack

Module	Technology
Agent Reasoning	Claude API
Tool Calls	MCP (Model Context Protocol)
Command Channel	SMTP / IMAP (email)
Runtime	Python

Repository

github.com/ramadanma/ran-agent