Claude Code has shipped a feature that sounds futuristic until you use it: your terminal sessions can now send messages to each other. Rolling out in Claude Code v2.1.224 on macOS and Linux, cross-session messaging lets one session hand off a finding, ask a question, or coordinate work with another session, without you re-explaining the context in both places.
The mechanism matters as much as the feature. Claude sends a summary of what you want to convey, not your raw history or files, so the receiving session gets a digest instead of your entire context window. Anthropic describes the sweet spots as handing over findings, coordinating parallel worktrees, checking status on long-running jobs, and replying from another machine.
How It Works
Instead of manually pasting context between terminals, you ask Claude to pass a message to a specific session. Claude composes the message itself; you supply the content you want to get across, and it handles the wording. The receiving session picks it up mid-task, so a change in one session that breaks what another is building can trigger a warning before you even notice.
The design deliberately keeps boundaries in place. Cross-session messaging does not approve permission requests or change configuration. Commands like /compact arrive as plain text rather than being executed, and if acting on a received message would require a permission, the receiving session still asks you for approval. So this is a communication channel, not a remote-control channel.
Why It Matters for Real Workflows
If you run multiple Claude Code sessions against the same project, one working on the frontend, one on the backend, one babysitting a long migration. The coordination overhead used to be on you. Now it is partially delegated: when one session resolves a question another is blocked on, it can send the answer across, and when a change breaks a dependency, the dependent session gets warned early instead of failing late.
That is a practical step toward multi-agent development inside a single tool, and it is worth watching how far Anthropic takes it. Our guide on using Claude Code for QA automation covers the kind of parallel workflows this unlocks.
The Caveats
A few honest limits. First, the feature is new and rolling out now; expect rough edges and version-specific behavior on Linux versus macOS. Second, because Claude composes the summaries, control over exactly what leaves your machine is indirect: you approve the content, but the encoding is the model’s. Third, the no-permissions rule means the receiving session still needs you at the keyboard for anything sensitive, so this automates communication, not authorization. If you rely on strict session isolation for security-sensitive work, test the summary scope first.
Related: see our Claude Code guide hub