Agent Gossip: a way for AI coding agents to talk to each other

AI coding agents are useful when they can focus on a task. But real software work rarely lives in one place. A product can span multiple repositories, tools, and machines. When agents work in those separate spaces, sharing context becomes part of the job.

I created Agent Gossip to explore that problem. The npm package is described as a way to let AI coding agents talk to one another across repos, tools, and machines. It includes a command line interface and can be used as an MCP server, so agent workflows can exchange information through the tools they already use.

Why agent communication matters

Imagine one agent working on an API and another updating the interface that consumes it. A change in one place can affect the other. Giving those agents a channel to ask questions, share updates, and coordinate can reduce the manual work of copying context from one session to another.

Agent Gossip is a small piece of infrastructure for that kind of collaboration. The published package exposes the gossip and agent-gossip commands. The related npm packages include shared message types and a WebSocket relay hub.

What I’m building toward

My interest is in making multi-agent work feel more like a connected team: each agent can keep its own task and environment while still passing along the information another agent needs. Agent Gossip is an experiment in giving that exchange a practical, reusable home.

If you are building with multiple AI coding agents, take a look at the package and the source. I’d be interested to hear where communication between agents creates friction in your workflow.