MCP channel
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MCP channelSend a message to the backend session asking for the API documentation"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP channel
A lightweight MCP channel server that lets agentic agents discover and communicate with each other through direct messages and named channels. Built for multi-agent orchestration.
Architecture
┌─────────────┐ HTTP/SSE ┌────────────┐ HTTP/SSE ┌─────────────┐
│ Agent │◄────────────────►│ Broker │◄────────────────►│ Agent │
│ Session A │ (channel.ts) │ (broket.ts)│ (channel.ts) │ Session B │
└─────────────┘ └────────────┘ └─────────────┘Each agent session spawns its own channel.ts MCP server as a subprocess. These servers connect to a shared broker.ts HTTP server that routes messages between them.
Related MCP server: Agent Team MCP Server
Quick Start
1. Install dependencies
bun install2. Start the broker
The broker must be running before any agent sessions connect:
bun run broker.tsBy default it listens on port 8900. Override with CHANNEL_BROKER_PORT=9000 bun run broker.ts.
3. Start agent sessions
In separate terminals, start agent sessions with different names. For Claude Code:
# Terminal 1
CHANNEL_SESSION_NAME=frontend claude --dangerously-load-development-channels server:channel
# Terminal 2
CHANNEL_SESSION_NAME=backend claude --dangerously-load-development-channels server:channel
# Terminal 3
CHANNEL_SESSION_NAME=coordinator claude --dangerously-load-development-channels server:channelEach session auto-registers with the broker and can immediately communicate.
Tools Available to the agent
Tool | Description |
| Send a direct message to a session (by name) or to a channel |
| List all registered sessions and their online status |
| List available channels and their members |
| Create a named channel for group communication |
| Join an existing channel |
| Leave a channel |
Usage Examples
Direct messaging
Tell one agent session:
"Send a message to the backend session asking it to create the API endpoint for /users"
Channel-based coordination
Tell the coordinator session:
"Create a channel called 'project' and invite the other sessions to coordinate on building the user dashboard"
Multi-agent workflow
Start a "coordinator" session and tell it the overall plan
Start "frontend" and "backend" sessions
The coordinator creates a channel, and instructs each agent on their role
Agents communicate through the channel to coordinate their work
Configuration
Environment Variable | Default | Description |
|
| Human-readable name for this session |
|
| Broker server URL |
|
| Port for the broker (broker.ts only) |
How It Works
Broker (
broker.ts): A lightweight HTTP server that maintains a registry of sessions, manages named channels, and routes messages via SSE streamsChannel MCP Server (
channel.ts): An MCP server that agent spawns as a subprocess. It registers with the broker, opens an SSE connection for inbound messages, and exposes tools for sending messagesMessage delivery: When the agent calls
send_message, the MCP server POSTs to the broker, which pushes the message via SSE to the recipient's MCP server, which delivers it to the recipient's agent session as a<channel>notification
This server cannot be deployed
Maintenance
Related MCP Connectors
End-to-end encrypted messaging and work coordination for autonomous AI agents.
Agent communication platform for agent to agent messaging via MCP. Messages, channels, skills.
Persistent discussions, shared context and subscriptions for independently operated AI agents.
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceA real-time inter-agent switchboard, delivered as one centralized streamable-HTTP MCP server. Any MCP-capable agent can message, coordinate, and stay ambiently aware of others.AGPL 3.0
- AlicenseNot gradedqualityDmaintenanceEnables real-time communication and orchestration of multiple AI agents with a web dashboard for monitoring agent activities, tasks, and artifacts.MIT
- AlicenseNot gradedqualityBmaintenanceA plugin-based MCP coordination server for AI agents, providing chat channels, agent registry, and event bus with ACID guarantees and zero infrastructure.1MIT
- AlicenseNot gradedqualityBmaintenanceA local message broker for AI CLI agents, enabling multiple agent sessions to communicate, send tasks, and share results through a single HTTP endpoint.MIT