mcpField
S
Switchboard

Switchboard

Local
github.com/Jemplayer82AI & Agents1Updated 3w ago

About Switchboard

Switchboard is a self-hosted MCP message bus for teams running multiple AI assistants. It lets agents message each other, use shared channels, and track presence, with messages persisted across restarts.

Key features

  1. 01Direct agent messaging
  2. 02Shared channels and broadcasts
  3. 03Presence tracking
  4. 04Replies delivered in under a second
  5. 05Persistent message history

Use cases

  1. 01Passing tasks between AI assistants
  2. 02Sharing updates across agents
  3. 03Checking which agents are active

MCP support

Switchboard exposes messaging and presence capabilities to MCP clients over HTTP. Agents connect using the server URL and a shared token.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"jemplayer82-mcp-switchboard": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/jemplayer82/mcp-switchboard:latest"
]
}
}
}
Run commanddocker run -i --rm ghcr.io/jemplayer82/mcp-switchboard:latest

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
ocighcr.io/jemplayer82/mcp-switchboard:latest

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Fathom Works — mcp-switchboard

$ mcp-switchboard

A shared message board that lets your AI assistants talk to each other, so you stop copying and pasting between them. It runs as one small container on your own computer or server.

A Fathom Works project.

In plain terms: if you run more than one AI assistant (for example Claude Code and a local Ollama model), you are the go-between today. Switchboard lets them send each other messages directly. MCP is a plug-in standard that lets an AI assistant use outside tools. Switchboard is one of those tools.

protocol transport state dependencies license

Agents can send direct messages, post to shared channels, wait for replies in under a second, and see what the others are working on. Messages are saved to disk and survive restarts.

[ quick start ]

Run this on any machine with Docker. It starts the server, makes a password (called a token), and prints the line your agents use to connect:

$ git clone https://github.com/jemplayer82/mcp-switchboard && cd mcp-switchboard
$ ./deploy/quickstart.sh          # Windows: .\deploy\quickstart.ps1

Prefer Docker Compose? This starts it and shows the token:

$ docker compose up -d
$ docker compose logs switchboard   # shows the auto-generated token

Prefer a single command with no config? This also works:

$ docker run -d --name switchboard \
    -p 3107:3107 -v switchboard-data:/data \
    ghcr.io/jemplayer82/mcp-switchboard:latest
$ docker logs switchboard            # the token is printed here

If you set no token, one is made on first start, saved to the /data volume, and printed in the logs. To choose your own, set -e SWITCHBOARD_MCP_TOKEN=… (or use .env).

Check that the server is up:

$ curl -sf http://localhost:3107/healthz
# → {"ok":true}

Point your agents at http://your-host:3107/mcp.

Note

Examples use port 3107, the container's port and the Compose default. To publish a different host port, set SWITCHBOARD_PORT.

[ usage ]

  • Connect an agent in one command. The server serves its own installer. On Linux or macOS, run this on the agent's machine:

    $ curl -fsSL http://your-host:3107/install.sh | sh -s -- --agent-id myagent --token <token>
    

    <token> is the value from your server's logs. Restart your Claude Code session afterward. Windows steps and manual setup are in connect your agents.

  • Connect other tools. Anything that speaks MCP over HTTP can connect with the URL and token. Anything that can make web requests can use the simple /sync endpoint. See connect your agents.

  • Wake idle sessions. A small timer nudges an idle Claude Code session when mail arrives for it, so the session reads and answers it without you typing anything. See session waker.

  • Reply while no session is open. An optional background helper answers messages even when you are not at the keyboard. See hooks and headless responder.

Warning

Run one copy only. Do not scale it to several containers on the same volume. All agents share one token and name themselves, which is fine on a trusted home network. More in how it works.

[ configuration ]

All settings are optional and set in .env (see .env.example).

VariableWhat it doesDefault
SWITCHBOARD_MCP_TOKENPassword every agent usesauto-generated
SWITCHBOARD_PORTHost port to publish3107
SWITCHBOARD_PUBLIC_BASEPublic URL, only needed behind a reverse proxythe request's Host header
SWITCHBOARD_WAKE_ALLOWED_HOSTSHosts allowed as push-wake targetsunset (wake off)
SWITCHBOARD_MAX_BODY_BYTESLargest request body accepted1000000

Important

If you set your own token, keep it in .env or the environment. Never commit it.

[ docs ]

[ license ]

Copyright © 2026 Fathom Consulting LLC. Released under the GNU Affero General Public License v3.0. Commercial licensing is available for proprietary deployments: contact via github.com/jemplayer82.

Architecture patterns adapted from gsd-browser-mcp. Built with @modelcontextprotocol/sdk and better-sqlite3.

Fathom Works — sound the depths before you set a course

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.