1300 166 138 Tell us what you need
News / Anthropic

Claude Code sessions can now send each other messages

By Monika Tantau • Published 3 months ago
Claude Code sessions can now send each other messages

Key takeaways

• Claude Code v2.1.224 adds two tools, ListAgents and SendMessage, so sessions can message each other
• Only plain text crosses between sessions, never conversation history or files
• On the same machine the message travels over a local socket and never reaches Anthropic servers
• A message cannot approve a permission or change configuration in the receiving session

Most Read

Anthropic has added cross-session messaging to Claude Code, letting one terminal session send a message to another running on the same machine. The feature shipped in version 2.1.224 and runs on macOS and Linux, according to 9to5Mac. It is on by default, with nothing to enable.

The use case is narrow and practical. Developers increasingly run several Claude Code sessions at once, each on a different part of a codebase, a working pattern most coding AI tools still treat as one session at a time. Until now, when one session made a change that broke what another was building, the developer relayed it by hand between terminals.

What the two new tools do

Claude Code now ships two tools the model calls on its own: ListAgents, which finds the other sessions it can reach, and SendMessage, which delivers text to one of them by name. Developers never invoke either directly.

A message is plain text one Claude writes to another. Anthropic’s documentation is explicit that it carries never conversation history or files. To move a whole conversation, you still resume the session rather than message it.

Sessions answer to a name, set with the /rename command or a –name flag at launch. Without one, Claude Code derives the name from the working directory, so a session in a folder called myapp becomes something like myapp-3f.

Where the message actually travels

The route depends on where the other session runs. Between two sessions on the same machine, the message goes over its own socket on disk and never touches Anthropic’s servers.

Messages to a session on another machine do go through Anthropic, arriving over that machine’s Remote Control connection. That path allows replies only: a session can answer a message that came from another machine, but cannot start the exchange.

Sessions also have to see the same filesystem to find each other. A session inside a container and one on the host cannot connect, because each registers itself in files the other cannot read.

What a message cannot do

Anthropic has drawn the boundaries tightly. A message from another session does not count as user consent, so it cannot answer a pending permission prompt. The receiving Claude is instructed never to change permission settings, CLAUDE.md or other configuration because another session asked it to.

A slash command written into a message arrives as plain text and is never executed. If acting on the message needs a permission the receiving session does not have, the usual prompt still fires.

Loops are throttled. Claude Code rate limits repeated messages per sender, drops identical repeats arriving in quick succession, and caps unread messages at 50 per session, so two sessions cannot talk each other into an endless exchange.

Why Australian businesses should care

For Australian teams, the useful detail is the split in that routing table. Work that stays on one machine stays on that machine, which matters for firms handling material covered by the Privacy Act 1988 or client data they have contracted to keep onshore.

That distinction only holds if the path between machines is closed. Setting isolatePeerMachines to true forces an approval prompt before any message leaves the machine, and the setting can be switched on from a project file in version control but not off.

Administrators can disable the feature outright through managed settings, denying the SendMessage and ListAgents tools and setting crossSessionInbound to refuse. For teams handling sensitive material such as document automation, that is worth deciding before developers turn it on themselves.

Why Australian businesses should care

A first step toward agent handovers

The feature is deliberately small. It moves a sentence between two sessions a developer already started and steers, not a coordinated fleet, and Anthropic points to its agent teams feature for that.

It is also not available everywhere. Native Windows is out, and so are Amazon Bedrock, Claude Platform on AWS, Google Cloud’s Agent Platform and Microsoft Foundry.

The direction is clear enough all the same. As more work moves to local AI agents running in parallel, the harder question stops being how to prompt one agent well and becomes what one agent should be allowed to tell another.

Scroll to Top