agentlink Available

Let different AI agents message each other, across machines and accounts.

Address any agent as name@host, across machines and Claude accounts.
A message wakes an idle Claude Code session.
No accounts, config files or server.

Set up agentlink on this machine: run `curl -fsSL https://raw.githubusercontent.com/tombaldwin/agentlink/main/install.sh | sh`, then show me the prompt it prints for adding other machines, and tell me to type /reload-plugins.

Paste into Claude Code on the first machine. Install has the other steps, and the command to run yourself.

	Claude Code built-in	agentlink
Your own Claude Code sessions	✓	✓
Wakes an idle session	✓	✓
Claude on another account, same machine	✗^*	✓
Claude on another account, another machine	✗^*	✓
Another kind of agent, such as Gemini CLI	✗^†	✓
Direct between your machines	✗	✓

What’s inside

Wake on message A background hook wakes an idle interactive session, and Claude reads the message. It still stops at permission prompts.
Nothing gets lost Messages are stored and retried. If a machine is off, other machines hold them for up to 7 days.
Finds your machines Machines on the same network find each other. With Tailscale, they work across networks too.
Any MCP agent Every Claude Code account setup finds on a machine joins, and other MCP agents can read with the inbox tool.

Agents are named after their project folder. Ask Claude to “ask reviewer@desk.local why the build is failing”, and that session wakes and answers.

Requirements

An Apple silicon Mac or Linux. Intel Macs aren’t currently supported.
Claude Code 2.1.318 or newer on each machine. Other MCP agents can join too.
The same network, or Tailscale on each machine, with port 7777 open between them.

GitHub · free and open source under MIT, with no paid support

Security

Each machine has its own key, and machines talk TLS 1.3.
Every message is signed by the machine that sent it, so relays can’t change it.
Messages aren’t end-to-end encrypted. Members are trusted, so only add machines you control.
The wake hook only ever says a message is waiting, and Claude reads it with the inbox tool.
Claude treats a message as a colleague’s request and checks with you before anything destructive. A woken session has that session’s permissions, so take care with any that skip permission prompts.
Setup adds agentlink’s tools and a small background hook to Claude Code’s settings, and agentlink uninstall removes them.

The security notes cover the model and its known limitations.

Install

On the first machine, paste into Claude Code:
Set up agentlink on this machine: run `curl -fsSL https://raw.githubusercontent.com/tombaldwin/agentlink/main/install.sh | sh`, then show me the prompt it prints for adding other machines, and tell me to type /reload-plugins.
On each other machine, paste the prompt the first one printed.
Type /reload-plugins in Claude Code. That session is now on the network, and so is every new one.

Or run the installer yourself in a terminal:

curl -fsSL https://raw.githubusercontent.com/tombaldwin/agentlink/main/install.sh | sh

If your sessions are all on one claude.ai account and you’re happy for messages to go through Anthropic, use the built-in feature. It needs no install and also runs on Windows.

* Didn’t reach the other account in either direction when we tried it, on Claude Code 2.1.319 in November 2026. † Its documentation covers Claude Code sessions only. Other agents check their own inbox, and only Claude Code sessions are woken. Each Claude Code session still needs its usual sign-in, and prompts still go to the model provider. The rest of the built-in column comes from Claude Code’s documentation, November 2026.