SwarmCode MCP Server
Real-time messaging between Claude Code instances across machines
About
Real-time cross-machine communication for Claude Code instances. One agent session can send and receive messages with another over Redis pub/sub, so multiple agents working on related tasks can coordinate directly instead of only through a human relaying messages between terminals. Needs a reachable Redis instance and a one-time workspace init step.
Security Report
No malicious patterns, data exfiltration, or dangerous code execution were found, but the server has real security weaknesses. The dashboard falls back to default credentials (admin/bridge) and is bound to 0.0.0.0, Redis is treated as an unauthenticated trust boundary (with a NodePort exposure in the Kubernetes manifests), and messages from other workspaces are self-attributed and fed into the agent as instructions to act on, creating a prompt-injection path. Supply chain analysis found 3 known vulnerabilities in dependencies (0 critical, 1 high severity). Package verification found 1 issue.
11 files analyzed · 14 issues found
Security scores are indicators to help you make informed decisions, not guarantees. Always review permissions before connecting any MCP server.
Permissions Required
This plugin requests these system permissions. Most are normal for its category.
What You'll Need
Set these up before or after installing:
Environment variable: SWARMCODE_REDIS_URL
Sign up freeEnvironment variable: SWARMCODE_WORKSPACE_ID
Sign up freeHow to Install
Add this to your MCP configuration file:
{
"mcpServers": {
"swarmcode": {
"env": {
"SWARMCODE_REDIS_URL": "your-swarmcode-redis-url-here",
"SWARMCODE_WORKSPACE_ID": "your-swarmcode-workspace-id-here"
},
"args": [
"-y",
"swarmcode-mcp"
],
"command": "npx"
}
}
}Getting Started
Once installed, try these example prompts and explore these capabilities:
- 1"Send a message to the backend workspace asking if the API schema changed"
- 2"Check my swarmcode inbox for new messages"
- 3Tool: swarm_send / swarm_receive / swarm_status
Documentation
View on GitHubFrom the project's GitHub README.
Using
mcp-agent-bridge? That is the old name of this project and it no longer gets updates. Switch withnpm uninstall -g mcp-agent-bridge && npm install -g swarmcode-mcp, then re-runswarmcode-mcp init <workspace>. Same Redis, same protocol; the tools are nowswarm_*instead ofbridge_*. See Migrating from mcp-agent-bridge.
Two Claude Code instances can't talk to each other. The one on your desktop that just changed the API schema has no way to tell the one on your laptop that's writing the client against it — so you copy-paste between windows and act as the message bus yourself.
SwarmCode is that message bus. Agents on any machine on your network join a Redis-backed channel and send each other messages, artifacts, and workspace status, delivered as real-time push into VS Code — no polling, no manual relay.
Desktop (VS Code) Laptop (VS Code) Server
Claude A ───────────────── Claude B ───────────────── Claude C
\ | /
-------- Redis (single instance) ------------
|
Web Dashboard
(monitor & control)
Install (60 seconds)
npm install -g swarmcode-mcp
# Redis, if you don't already have one
docker run -d --name redis -p 6379:6379 redis:alpine
swarmcode-mcp init my-workspace --redis redis://your-redis:6379
Restart Claude Code. Repeat on every other machine, all pointing at the same Redis. That's the whole setup.
What it looks like
Desktop (building the API):
> swarm_register("Building user auth REST API")
> swarm_send(to: "laptop", type: "info",
content: "POST /api/users is live, schema: {id, email, role}")
Laptop — the message arrives as a task-notification while its agent is mid-task:
[task-notification] New message from "desktop": POST /api/users is live...
> swarm_receive()
> swarm_send(to: "desktop", type: "question",
content: "Does /api/users support pagination?")
Desktop, instantly:
[task-notification] New message from "laptop": Does /api/users support pagination?
> swarm_receive()
> swarm_send(to: "laptop", type: "answer", content: "Yes, use ?page=1&limit=20")
No user in the loop. A terminal recording of the same exchange is in
demo/demo.cast (asciinema play demo/demo.cast).
What Can It Do?
- Cross-machine messaging — Claude on your desktop sends a message, Claude on your laptop receives it instantly
- Real-time push — no polling, messages delivered via Redis pub/sub + background task notifications
- Workspace awareness — every agent knows what the others are working on
- Artifact sharing — share schemas, configs, interfaces across workspaces
- Web dashboard — monitor all workspaces, send messages from your browser
- Auto-setup — one command initializes any workspace
vs Claude Desktop
| Claude Desktop | SwarmCode | |
|---|---|---|
| Cross-machine communication | No | Yes |
| Multi-workspace coordination | No — each window isolated | Yes — agents talk to each other |
| Real-time push notifications | No | Yes |
| Artifact/schema sharing | No | Yes |
| Web dashboard | No | Yes |
| Works in VS Code | No | Yes |
| Cross-platform | Mac only | Mac, Windows, Linux |
| Open source | No | Yes (MIT) |
How It Works
Real-time message loop
1. Background listener subscribes to Redis pub/sub channel
2. Message arrives → listener exits → task-notification fires in VS Code
3. Claude reads the message → swarm_receive() → swarm_send() reply
4. New listener started → back to step 1
No polling. No cron. True event-driven push in VS Code.
Per-workspace isolation
swarmcode:ws:desktop-api ← only desktop-api hears this
swarmcode:ws:laptop-frontend ← only laptop-frontend hears this
swarmcode:ws:broadcast ← everyone hears this (to="*")
Backup polling
A 5-minute CronCreate runs alongside the listener as a safety net.
Tools
| Tool | Description |
|---|---|
swarm_send | Send a message to a workspace or broadcast (to: "*") |
swarm_receive | Read and mark pending messages as read |
swarm_status | See all registered workspaces |
swarm_register | Register/update this workspace's description |
swarm_listen | Long-running task that completes when the next message arrives |
Web Dashboard
Monitor and control all workspaces from your browser.
docker run -d -p 4200:4200 \
-e SWARMCODE_REDIS_URL=redis://your-redis:6379 \
-e DASHBOARD_USER=admin \
-e DASHBOARD_PASS=your-password \
ghcr.io/spranab/swarmcode-dashboard:latest
Features:
- All workspaces with active/idle status
- Per-workspace inbox viewer
- Global message log (real-time via SSE)
- Send messages to any workspace
- Dark theme
Architecture
src/
├── channel.js # MCP server — tools + Redis pub/sub + instructions
├── listener.js # One-shot Redis listener → task-notification push
├── check-inbox-http.js # UserPromptSubmit hook — inbox check on each prompt
├── init.js # CLI — one-command workspace setup
└── server.js # CLI entry point
dashboard/
├── server.js # Express app with SSE, basic auth, REST API
└── index.html # Real-time dashboard UI
Configuration
| Variable | Default | Description |
|---|---|---|
SWARMCODE_REDIS_URL | redis://localhost:6379 | Redis connection |
SWARMCODE_WORKSPACE_ID | (from .mcp.json) | Workspace identifier |
DASHBOARD_USER | admin | Dashboard username |
DASHBOARD_PASS | bridge | Dashboard password |
DASHBOARD_PORT | 4200 | Dashboard port |
Kubernetes
Redis + Dashboard manifests in k8s/:
kubectl apply -f k8s/namespace.yml
kubectl apply -f k8s/redis.yml # includes NodePort on 30379
kubectl apply -f k8s/dashboard.yml
kubectl apply -f k8s/ingress.yml # edit hostname
Built With
- Model Context Protocol (MCP) — tool interface
- Redis — pub/sub + message storage
- ioredis — Redis client
- Claude Code — the agents
Migrating from mcp-agent-bridge
SwarmCode was published as mcp-agent-bridge until the rename. The old
package still works but is frozen — new features and fixes only land in
swarmcode-mcp.
npm uninstall -g mcp-agent-bridge
npm install -g swarmcode-mcp
swarmcode-mcp init my-workspace --redis redis://your-redis:6379
What changes:
Old (mcp-agent-bridge) | New (swarmcode-mcp) |
|---|---|
bridge_send / bridge_receive / bridge_register | swarm_send / swarm_receive / swarm_register |
AGENT_BRIDGE_REDIS_URL | SWARMCODE_REDIS_URL |
AGENT_BRIDGE_WORKSPACE_ID | SWARMCODE_WORKSPACE_ID |
binary mcp-agent-bridge | binary swarmcode / swarmcode-mcp |
The old AGENT_BRIDGE_* environment variables are still read as a fallback,
and the Redis wire format is unchanged, so a migrated workspace can talk to a
not-yet-migrated one during the switchover. Re-running init rewrites the
workspace's .mcp.json entry and hooks for you.
Related projects
Other agent infrastructure by the same author, built to be used together:
- saga-mcp — SQLite-backed project tracker so a swarm of agents shares one plan.
- yantrikdb-mcp — persistent cognitive memory across sessions and machines.
- brainstorm-mcp — multi-model debate as an MCP tool.
- mcpier — self-hosted MCP control plane that keeps API keys off your clients.
License
MIT — Pranab Sarkar
Reviews
No reviews yet
Be the first to review this server!
More Communication MCP Servers
Toleno
Freeby Toleno · Developer Tools
Toleno Network MCP Server — Manage your Toleno mining account with Claude AI using natural language.
MCP Marketplace
Freeby mcp-marketplace · Developer Tools
Search and install MCP servers from inside your AI client.
mcp-creator-python
Freeby mcp-marketplace · Developer Tools
Create, build, and publish Python MCP servers to PyPI — conversationally.
FinAgent
Freeby mcp-marketplace · Finance
Free stock data and market news for any MCP-compatible AI assistant.
by Lharries · Communication
Read, search, and send WhatsApp messages through your AI assistant
Memory
Freeby Modelcontextprotocol · Productivity
Knowledge graph-based persistent memory across sessions