I usually have two or three Claude Code sessions open at once. One in the backend repo, one in the frontend, sometimes a third doing infra. Half my day was alt-tabbing between them. I'd be in the frontend, hit "wait what shape did the user object end up as", switch to the backend session, ask, copy the answer, switch back, paste. The other Claude already knew the answer. I was the slow part of the loop

So I wrote a plugin. It's called Relay. In the frontend session I just type something like:

  ask the backend session what the user object looks like

The backend session picks up the question between turns and writes a reply. The reply shows up in my frontend session as a notification a few seconds later. Broadcasts work the same way — "ask everyone what they're working on" goes to every session and replies come back as they finish

Mechanism: Claude Code has a channels capability that lets an MCP server push a message into a session between turns. Relay uses that. Every session runs a small MCP server, and a single hub daemon on your machine routes messages between them over a Unix socket. The first session you start spawns the hub; it shuts itself down about five minutes after the last session disconnects. Same-host only, no auth, no network calls

Worth saying what this isn't. There are a few projects in the "make Claudes coordinate" space and most of them are orchestration frameworks where one Claude is the boss and the others are workers. Relay isn't that. It doesn't tell your sessions what to do or split tasks for you. It just lets the sessions you already have open ask each other questions, the same way you'd ping a coworker

Repo and install steps: https://github.com/innestic/claude-relay (MIT)

If you already run multi-session workflows, I'd love to hear what your setup looks like and where it gets annoying. That's the part I want to figure out next