AgentLink

English | 中文

Access your local Claude Code agent from anywhere — through a shareable web URL.

AgentLink runs a lightweight agent on your dev machine, connects it to a cloud relay, and gives you a clean chat interface at https://msclaude.ai/xxxx. All code execution happens locally. The relay just forwards encrypted messages.

Browser (any device)  ←→  Cloud Relay  ←→  Local Agent  ←→  Your Codebase

Quick Start

# Install
npm install -g @agent-link/agent

# Start (prints a session URL)
agentlink-client start --daemon

# Open the URL in any browser — done.

Three Modes

Chat

Remote pair-programming with Claude Code. Send prompts from any browser, get real-time streaming responses with tool calls, markdown, and code blocks rendered live.

Teams

Orchestrate multiple Claude agents in parallel. A Lead agent plans and delegates tasks to subagents, then synthesizes results — with a live kanban dashboard tracking progress.

Loop

Schedule recurring prompts with cron expressions. Automate code reviews, dependency checks, report generation, and other routine tasks that run on your local codebase.

Why AgentLink?

• Access from anywhere. Phone, tablet, another machine — just open the URL.
• Everything runs locally. Your files never leave your machine. The relay only sees encrypted messages.
• No setup on the browser side. No extensions, no local server, no port forwarding. One URL.
• Session history. Resume previous conversations. Your Claude session files are read directly.
• Real-time streaming. Tool calls, markdown, code blocks — all rendered live as Claude works.
• Multi-agent Teams. Spawn parallel Claude subagents from the browser. A Lead agent plans, delegates tasks, and synthesizes results — with a live kanban dashboard.
• End-to-end encryption. All WebSocket traffic is encrypted with XSalsa20-Poly1305 (TweetNaCl).

How It Works

1. agentlink-client start connects to the relay server via WebSocket.
2. The server assigns a unique session URL (96-bit random ID).
3. You open the URL in a browser. The browser connects to the same relay.
4. Messages flow: browser → relay → agent → Claude CLI → agent → relay → browser.
5. Claude runs with --permission-mode bypassPermissions, so it can work autonomously.
6. All messages between relay↔agent and relay↔browser are encrypted with per-session keys.

Commands

# Agent
agentlink-client start [--daemon] [--server URL] [--name NAME] [--dir PATH] [--password PWD]
agentlink-client stop
agentlink-client status
agentlink-client config list|get|set
agentlink-client service install|uninstall
agentlink-client upgrade

# Self-hosted relay server
agentlink-server start [--daemon] [--port PORT]
agentlink-server stop
agentlink-server status

Self-Hosting

You can run your own relay server instead of using the public one:

npm install -g @agent-link/server
agentlink-server start --daemon --port 3456

# Then point agents to your server
agentlink-client start --daemon --server ws://your-server:3456

Requirements

• Node.js 18+
• Claude CLI installed and authenticated (claude command available)

Architecture

Component	Tech
Agent	TypeScript, Commander.js, ws
Server	Express, ws, modular architecture (SessionManager, AuthManager, MessageRelay)
Web UI	Vue 3 SFC + Vite, marked.js, highlight.js
Encryption	TweetNaCl (XSalsa20-Poly1305 secretbox)

Links

• npm: @agent-link/agent
• npm: @agent-link/server
• GitHub

License

MIT