- Published on
Claude Code Cross-Session Messaging: Your AI Agents Can Now Talk to Each Other
Just as a developer messages a colleague on Slack, Claude Code sessions now send messages directly to each other.
Think about how software teams work. When one developer changes an API spec, another developer who doesn't know about it ends up with silently broken code. Human developers send a quick Slack message: "Hey, I changed the API — check your end." But AI agents? Until now, they each worked in isolation.
Starting with Claude Code v2.1.224 (released August 2026), that pattern has changed. Cross-Session Messaging allows independent Claude Code sessions running simultaneously to message each other directly. The September update expanded support to native Windows (v2.1.234 or later) alongside macOS, Linux, and WSL2.
How It Works
Two core tools make this possible.
ListAgents — Returns a list of sessions reachable from the current machine. Terminal sessions, VS Code extensions, and web sessions all appear here. Each row shows the session name, type, and status (busy or idle).
SendMessage — Specify the target session name from ListAgents in the to field, and put the content in message. The receiving session picks this up on its next turn.
Consider a real scenario where two sessions are modifying the same repository simultaneously:
- Session A: Changing the signature of
processPayment()in the billing module - Session B: Writing test code that calls the payment API
The moment Session A changes the function signature, it fires a SendMessage to Session B: "Payment function signature changed. New parameters need checking." Session B learns this before piling more code on top of a broken foundation.
Why This Matters
Reducing the "coordination tax" in parallel development
The hardest problem in agent-based development is conflicts that arise when multiple agents work simultaneously. Until now, conflicts were always discovered after the fact — in PR review, CI failures, or runtime errors.
Cross-session messaging moves that timeline earlier. The moment a change occurs, affected sessions can be notified immediately.
The first step toward an agent network
In a broader context, this feature signals that Claude Code''s multi-agent architecture is maturing. Sessions are no longer isolated execution units — they become nodes in a network that can share state and coordinate.
It''s plain text messages for now, but the foundation is laid for patterns where one session waits on another''s completion, or receives results to inform its own decisions.
Practical Tips
1. Prevent conflicts on shared repositories When multiple agents are editing the same codebase, broadcast interface changes or shared utility modifications via SendMessage before the other session touches the same files.
2. Share progress on long-running tasks When a build, test run, or data processing job finishes, send a completion signal to sessions waiting on that result. Build event-driven flows without polling.
3. Short-circuit error propagation When an unexpected error occurs in one session, send early warnings to downstream sessions that would be affected. Stop cascading failures before they spread.
Supported Platforms
| Platform | Minimum Version |
|---|---|
| macOS | v2.1.224 |
| Linux | v2.1.224 |
| WSL2 | v2.1.224 |
| Windows (native) | v2.1.234 |
Closing Thoughts
The shift from "AI working alone" to "AI working together" is accelerating. Cross-session messaging is an indispensable block in that transition.
Beyond writing code in natural language, agents now recognize and coordinate each other''s work. Just as a software team communicates via Slack, Claude Code sessions exchange messages. The next chapter of humans and AI working together is opening.
Source: