Claude Code v2.1.224｜セッション間 SendMessage が解禁｜毎日Changelog解説Claude Code v2.1.224 enables cross-session communication via SendMessage,…

Anthropic's Claude Code has added a mechanism for independently launched sessions to talk to one another directly, a change that lands in version v2.1.224 and expands how developers can build multi-agent workflows. The update introduces two capabilities working in tandem: SendMessage, which lets one running session pass a message straight to another, and ListAgents, which lets a session discover the other agents it can reach. Together they turn what were previously isolated instances into participants that can coordinate.

Until now, running multiple Claude Code sessions typically meant each one operated in its own silo, with no built-in channel for one session to hand work to, or request information from, another. Coordinating them often relied on external glue such as shared files, scripts, or a human copying context between terminals. With SendMessage, a session can address a specific counterpart and exchange messages as part of an ongoing task. ListAgents provides the discovery layer, returning the set of available agents so a session can identify who is present before sending anything. This pairing of discovery and messaging is the foundational pattern most agent-to-agent systems rely on.

The practical effect is that multi-agent coordination becomes more flexible. Developers can now design arrangements where, for example, one session focuses on a particular part of a codebase while another handles tests or documentation, with the two communicating results and requests directly rather than through a manual handoff. The source excerpt indicates the feature is oriented around finding and reaching a counterpart session, and it hints that discovery may not be limited strictly to sessions started in the same context; because that portion of the changelog text is truncated, the exact scope of cross-machine or cross-environment reach is not fully clear from the excerpt and is best treated cautiously until the complete documentation is consulted.

This move fits within a broader direction Claude Code has been taking toward agentic and modular workflows. The tool already supports subagents, which let a primary agent delegate scoped tasks to helper agents within a single session, and it integrates with the Model Context Protocol (MCP), an open standard for connecting AI assistants to external data sources and tools. Cross-session SendMessage appears to complement those features by extending communication beyond a single process boundary, so that separately started sessions, rather than only nested subagents, can participate in a coordinated effort. For teams already orchestrating several terminals or automated runs, a native messaging primitive reduces the need for bespoke inter-process plumbing.

It is worth situating this against the wider industry interest in multi-agent orchestration. Frameworks and patterns for having multiple language-model agents collaborate, negotiate, or divide labor have become a common theme across developer tooling, and standardized ways for agents to advertise their presence and exchange messages are a recurring building block. Anthropic building such primitives directly into Claude Code lowers the barrier for developers who want to experiment with these patterns without assembling their own coordination infrastructure.

The version number itself, v2.1.224, and the fact that this was surfaced through a daily changelog translation effort, reflect Claude Code's rapid release cadence. The Qiita-based source that reported this update translates and explains Anthropic's official changelog shortly after each release, which is one way the community tracks the tool's frequent incremental changes. Because these releases arrive so often, individual entries like this one tend to be narrowly scoped feature additions rather than sweeping overhauls, and their real-world impact usually becomes clearer as developers put them into practice.

For anyone considering adopting cross-session messaging, a few prerequisites are likely relevant: running more than one Claude Code session, understanding how ListAgents surfaces reachable agents, and designing task boundaries so that message passing adds value rather than complexity. As with any new agent-coordination capability, careful attention to how sessions address one another and what information they share will matter, particularly in setups where multiple agents can act on the same codebase. The feature broadens what is possible, but the concrete benefits will depend on how thoughtfully these interacting sessions are structured.