Claude Code cross-session messaging, extended to Codex, Kimi Code and ZCode

What Claude Code's built-in messaging does

From Claude Code v2.1.224 on macOS and Linux (including WSL 2), and v2.1.234 on native Windows, a Claude Code session can list your other sessions and send them a message with its ListAgents and SendMessage tools. Claude can do this on its own, for example to warn another session about a breaking change, or when you ask it to. Anthropic's documentation describes a message as

"a piece of text one Claude writes to another"

On supported versions it is on by default and needs no separate setup for sessions on the same machine; those messages are delivered through a per-session socket and never through Anthropic's servers. Delivery still depends on the receiving session being available and on its inbound-message settings. Source: Message your other Claude Code sessions (Claude Code Docs).

Where it stops

• Claude Code sessions only. It can reach subagents, agent-team teammates, your other local Claude Code sessions, cloud sessions and Remote Control sessions on other machines. A Codex, Kimi Code or ZCode session is not on the list.
• Other machines go through Anthropic. Messaging a session on another computer works when the sending session is connected to Remote Control (which needs a claude.ai sign-in) and the target session is running with Remote Control too; the message is relayed through Anthropic's servers. Sessions that authenticate with an API key or through providers such as Amazon Bedrock can still message sessions on the same machine, but cannot discover sessions beyond it.
• Text only. A message carries what Claude writes, not the sender's history or files. The receiving Claude works with its own files and permissions.

What the Codex plugin for Claude Code does

OpenAI's Codex plugin for Claude Code adds /codex:review, /codex:adversarial-review, /codex:rescue, /codex:transfer and job-management commands, so a Claude Code session can ask Codex for a review, delegate implementation work, run it in the background and collect the result. /codex:transfer creates a persistent Codex thread by importing the current Claude Code transcript, which you can then continue in the Codex app or terminal. It needs Node.js 18.18+ and a ChatGPT plan (including Free) or an OpenAI API key. Its FAQ says it

"delegates through your local Codex CLI and Codex app server on the same machine"

Where it stops

• Delegation starts in Claude Code. The documented flow is Claude Code handing work to Codex; the plugin does not provide general two-way messaging between live sessions, and a transferred thread is a copy, not a synchronized session.
• One machine. It uses the local Codex install and checkout.
• Two vendors. Kimi Code, ZCode and others are out of scope.

Side by side

Which one should you use?

• All your sessions are Claude Code: use the built-in messaging. It is already there.
• You want Codex to review or rescue work from inside Claude Code, on one machine: the Codex plugin is the simplest path.
• You run several vendors and want them to hand work to each other, or your sessions live on different Macs: use anytoany. It works alongside both. For open Claude Code sessions, anytoany wakes them through the same built-in inbox the official feature uses.

Set up anytoany

curl -fsSL https://raw.githubusercontent.com/Ericgood/any-to-any/main/install.sh | bash
anyd start
anyd list        # every addressable Claude Code / Codex / Kimi / ZCode session

Then, in Claude Code, say something like: "Ask the Codex session working on the API to rerun the route tests." To reach another Mac, run anyd pair --invite once; the two-Mac tutorial walks through it step by step. If your Claude Code sessions default to Bypass permissions, add "crossSessionInbound": "accept" to ~/.claude/settings.json so open sessions wake up as soon as a message arrives. Full details on the anytoany home page, in the GitHub README (supported agents and delivery paths) and in SECURITY.md (pairing, mutual TLS, trust boundary).

FAQ