开发者 yocontra 发布了名为 Pulsar 的模型上下文协议(MCP)服务器工具,旨在实现本地运行的多个 AI 编程代理之间的通信与消息传递 1。该工具将机器上的所有 AI 编码代理置于同一总线上,支持点对点、组播及广播消息模式 1。Pulsar 允许代理相互发现、标记状态并发送即时消息,以唤醒空闲代理或协调工作流 1。目前,Pulsar 支持的代理包括 Claude Code、Codex、OpenCode、pi、omp、Cursor 以及 Gemini CLI 1。
在安装部署方面,用户通过执行 npx pulsar setup 命令即可一键安装,系统会自动配置各代理环境 1。其底层技术基于一个小型守护进程,负责管理注册表和 SQLite 消息日志,并通过 Unix socket 与各代理交互 1。在安全性方面,Pulsar 默认将消息标记为来自对等代理而非用户,从而防止代理获得额外权限 1。此外,每条消息的大小限制为 64 KB,每分钟发送次数上限为 60 次 1。该工具强调无遥测收集,未来计划推出跨多台机器的代理通信功能,以及通过 Tailscale 实现的远程控制功能 1。
A developer known as yocontra has released Pulsar, a Model Context Protocol (MCP) server designed to facilitate communication and messaging between multiple locally running AI coding agents. 1 The tool functions as a central bus that allows agents such as Claude Code, Codex, OpenCode, pi, omp, Cursor, and Gemini CLI to discover each other, mark their status, and send instant messages to wake up idle agents or coordinate workflows. 1 By placing all AI coding agents on the same machine onto a unified bus, Pulsar supports point-to-point, multicast, and broadcast messaging capabilities. 1
The technical foundation of Pulsar relies on a small daemon process that manages a registry and an SQLite message log, interacting with individual agents through Unix sockets. 1 Installation is streamlined via a one-command setup using npx pulsar setup, which automatically configures the environments for supported agents. 1 To ensure security, the system marks messages as originating from peer agents rather than users, preventing any privilege escalation. 1 Additional safety measures include limiting each message to 64 KB and restricting sending frequency to 60 times per minute, while the tool emphasizes default secure settings and collects no telemetry data. 1
Looking ahead, the project plans to expand its functionality to support cross-machine agent communication and remote control features implemented through Tailscale. 1
评论
还没有评论,欢迎留下第一条。