Quick answer. Claude Code cross-session messaging: Anthropic has introduced cross-session messaging for Claude Code, allowing separate AI coding sessions to exchange information and coordinate work
Updated: 8 August 2026
Claude Code cross-session messaging is Anthropic’s new capability that lets separate Claude Code sessions talk to each other, as AI coding tools rapidly move beyond the idea of a single chatbot helping a developer write code.
Anthropic has introduced a significant new capability in Claude Code called cross-session messaging, which allows separate Claude Code sessions to communicate directly with one another.
In practical terms, you can have several Claude Code sessions working on different parts of a project at the same time – and those sessions can now exchange important information without you repeatedly copying and pasting instructions between them.
For developers already experimenting with parallel AI agents, this is a small-looking feature with potentially big consequences.
Table of contents
Key takeaways
- Anthropic has added cross-session messaging to Claude Code, letting separate AI coding sessions send each other short text messages to coordinate work.
- Only distilled text messages are shared – not full conversation history or files – so each session keeps its own context and permissions.
- It requires Claude Code 2.1.224+ and currently works on macOS and Linux (including WSL 2 on Windows); native Windows support is not yet available.
- Messages cannot approve permissions or change settings on another session’s behalf – the user stays in control of sensitive actions.
- It is a concrete step towards multi-agent AI development, where a developer manages a team of coordinating AI agents instead of one assistant.
What is Claude Code cross-session messaging?
Claude Code is Anthropic’s agentic coding environment, designed to work directly with software projects, execute commands, inspect code, edit files and perform development tasks.
Until now, running multiple Claude Code sessions often meant managing the communication between them yourself.
Imagine one Claude session working on your database while another session builds your API. If the database session changed a column name, you would normally need to tell the API session about that change.
With cross-session messaging, Claude can now do that itself. Anthropic explains that one Claude Code session can discover other available sessions and send them a message when information becomes relevant to their work.
For example, you could tell Claude:
“Explain what we just changed to the session working on the payments API.”
Claude then generates the appropriate message and sends it to the relevant session. Anthropic’s documentation says Claude may also decide to communicate with another session automatically when it recognises that the information is relevant.
It does not send your entire conversation
This is an important distinction. Claude Code does not simply dump the entire conversation from one session into another.
Anthropic says cross-session messaging sends text messages only. The receiving session does not automatically receive the other session’s conversation history or files.
That means Claude effectively distils the important information into a handover message. For example:
- Session A: Database migration
- Session B: API development
Session A might send something like:
“The database migration is complete. The new customer identifier is tenant_id, and it is now safe to rebase onto main.”
Session B receives that information while continuing its own work. That is much closer to two developers coordinating with each other than two isolated AI chat windows.
Why this Claude Code update matters
The biggest advantage is not necessarily that Claude can “talk to itself.” The real advantage is parallel work.
Developers increasingly use AI coding agents to perform several tasks simultaneously. You might have:
- one Claude Code session building the frontend
- another working on authentication
- another updating the database
- another running tests
- and another reviewing the code
Previously, the developer often became the communication layer between those agents. You had to remember which session knew what, explain changes repeatedly and manually move information between them.
Cross-session messaging reduces that overhead. Anthropic specifically highlights uses such as handing over findings, coordinating work across separate Git worktrees and reporting progress from long-running tasks.
Claude can message another session while it is working
One particularly interesting aspect is that the receiving Claude session does not necessarily have to finish its current task first.
According to Anthropic, incoming messages are read between tool calls during an active turn. This means an ongoing command is not suddenly interrupted, but Claude can receive the new information during the workflow and adjust what it does next.
If the receiving session is idle, the incoming message starts a new turn automatically. This makes cross-session messaging particularly useful for longer software projects where several agents may be operating concurrently.
Claude Code is moving towards multi-agent software development
Cross-session messaging also fits into a much larger shift taking place in AI development tools.
Claude Code already supports subagents, which are specialised Claude agents that can handle particular tasks in their own context windows and return the results to the main conversation. Anthropic also offers agent teams and background agents for more sophisticated parallel workflows – see our guide on why AI agent projects fail when treated like a chatbot for how to plan around this.
But cross-session messaging is slightly different. These are not necessarily agents created by one supervising Claude session. They can simply be independent Claude Code sessions that you started yourself.
That gives developers considerably more flexibility. You could effectively operate several specialised AI developers simultaneously:
Frontend Claude → Backend Claude → Testing Claude → Documentation Claude
Each could concentrate on its own part of the project while exchanging important information when required. For solo developers and small businesses, this starts to resemble having a small AI development team available on demand.
How Claude Code sessions communicate
Behind the scenes, Claude Code uses two tools. ListAgents allows Claude to discover other sessions it can communicate with. SendMessage allows Claude to send information to the chosen session.
The developer does not normally need to call these tools manually. Instead, you can simply ask Claude something like:
“Ask the session working on the database whether the migration has finished.”
Claude determines which session you mean and handles the communication. Users can also run:
/list-agents/peers
to see which Claude sessions are available for messaging. Sessions can also be given meaningful names using Claude Code’s /rename command or the --name option, making larger multi-session workflows easier to manage.
What about security and permissions?
Giving AI agents the ability to communicate raises an obvious question: can one Claude session tell another session to execute something dangerous?
Anthropic has added several safeguards. An incoming message from another Claude session is explicitly treated differently from an instruction coming directly from the user.
For example, another Claude session cannot use a message to approve a permission request on your behalf. It also cannot use the messaging system to change Claude Code configuration or permission settings.
Commands included inside a cross-session message are treated as ordinary text rather than automatically executed commands. If fulfilling a request requires additional permission, the receiving Claude session still needs to follow its own permission rules.
In other words, communication between agents does not remove the user’s authority over sensitive actions.
Messages between Claude sessions stay local where possible
There is another interesting architectural detail. When two Claude Code sessions are running on the same machine, Anthropic says the messages travel through a per-session local socket rather than through Anthropic’s servers.
Cross-machine communication is more restricted. Claude Code can also communicate with sessions connected through Remote Control or Claude Code on the web, but these currently operate primarily as reply channels rather than allowing a local session to initiate a new cross-machine conversation.
That distinction is particularly relevant for developers dealing with confidential or proprietary software.
Who can use Claude Code cross-session messaging?
There are currently some important limitations. Anthropic’s documentation states that cross-session messaging requires Claude Code version 2.1.224 or later.
At the time of writing, the feature is available on macOS and Linux, including Linux running through WSL 2 on Windows.
However, Anthropic says native Windows Claude Code does not currently support cross-session messaging. That is an important distinction for Windows developers. If you run Claude Code through WSL 2, however, the feature is supported.
There are also provider restrictions, including some Claude deployments through cloud platforms such as Amazon Bedrock, Google Cloud and Microsoft Foundry.
Example: building an application with four Claude Code sessions
Imagine you are building an online booking platform. Instead of asking one Claude session to handle everything, you could open four.
- Session 1 – Database: Handles the customer database, booking records and migrations.
- Session 2 – Backend: Builds APIs and business logic.
- Session 3 – Frontend: Creates the user interface.
- Session 4 – Testing: Continuously tests the application.
Now suppose the database Claude changes the booking table. Instead of you switching terminals and explaining the change to the backend agent, Database Claude can notify Backend Claude.
Backend Claude can update the API. Backend Claude could then notify Testing Claude that the endpoints have changed.
The developer remains in control, but Claude handles some of the coordination. That is a fundamentally different workflow from simply asking an AI chatbot to generate snippets of code.
Claude Code vs traditional AI coding assistants
Most first-generation AI coding assistants followed a fairly simple model: Developer → AI → Code.
Agentic coding tools are increasingly moving towards: Developer → multiple AI agents → coordinated software development, a shift we also cover in our breakdown of the Codex context window increase to 272K.
Claude Code’s cross-session messaging makes that second model significantly more practical. Instead of every AI agent operating in isolation, agents can begin to exchange the information required to keep a larger project moving.
The change may sound minor, but coordination is one of the biggest challenges when running multiple AI agents simultaneously.
Why this matters for small businesses and solo developers
You don’t need to run a large software company to benefit from this – see our recent roundup of AI operations updates business teams should act on. For small businesses, entrepreneurs and independent developers using AI to create their own software, the ability to parallelise development can be extremely useful.
One Claude session could work on a feature while another:
- investigates a bug
- updates documentation
- reviews security
- runs tests
- or prepares a database migration
Instead of waiting for one job to finish before starting the next, multiple tasks can progress concurrently. The human developer increasingly becomes the project manager and decision-maker, while AI agents handle more of the implementation work.
Is this the beginning of AI development teams?
Possibly. It would be an exaggeration to suggest that Claude Code sessions communicating with each other suddenly replaces an experienced software development team.
Humans still need to define requirements, evaluate architecture, review changes, manage security and decide whether the software actually solves the intended problem.
But the direction is significant. AI coding is moving from one assistant helping one developer toward systems in which multiple specialised agents can coordinate work.
Claude Code cross-session messaging removes one of the practical barriers to that model: constantly having to explain to each AI agent what the others have done.
For developers building increasingly complex applications with AI, that could become one of Claude Code’s most useful capabilities.
Frequently asked questions
What is Claude Code cross-session messaging?
It’s a Claude Code feature that lets independent sessions send each other short text messages so they can coordinate work without you manually copying context between terminals.
Does cross-session messaging share my entire conversation with other sessions?
No. Only short text messages are sent. The receiving session does not automatically get the other session’s full conversation history or files.
Which platforms support Claude Code cross-session messaging?
It requires Claude Code version 2.1.224 or later and currently works on macOS and Linux, including Linux via WSL 2 on Windows. Native Windows Claude Code does not yet support it.
Can one Claude Code session control another session’s permissions?
No. Messages from another session are treated as plain text, not commands. They cannot approve permission requests or change configuration on your behalf.
Do cross-session messages leave my computer?
When two sessions run on the same machine, messages travel through a local socket rather than Anthropic’s servers. Cross-machine communication is more restricted.
Final thoughts
Anthropic’s new Claude Code cross-session messaging feature is not simply another convenience feature. It is another step towards multi-agent AI development.
Separate Claude sessions can now exchange relevant information, report progress and coordinate parallel work while maintaining their own context and permissions.
For developers already using several Claude Code windows or terminals at the same time, the immediate benefit is simple: less copying, less re-explaining and better coordination.
The bigger story, however, is what this suggests about the future of software development. The next generation of AI coding may not involve one AI assistant sitting beside a developer.
It may involve developers managing an entire team of AI agents. And increasingly, those agents will be able to talk to each other.
Stay in the loop
Get practical AI tips and product updates, no spam, unsubscribe any time.
