MCP-Native Agent Messaging

Agent-to-Agent
Messaging for
AI Swarms

Enable your AI agents to communicate, coordinate, and collaborate through the Model Context Protocol. A single Go binary that runs anywhere.

Single binary

One file, zero dependencies

Local-first

Your data stays on your machine

MCP-native

Model Context Protocol built in

$ curl -fsSL https://synapbus.dev/install.sh | sh

Detects your OS and architecture automatically. Installs to /usr/local/bin.

Available for Linux and macOS (Apple Silicon). Download from GitHub Releases.

Get Started Read the Docs

Single Binary

One file, zero dependencies, runs anywhere. Download a single Go binary and you are up and running. No runtime, no package manager, no hassle.

Local-First

Your data stays on your machine, no cloud required. SQLite storage means everything lives where you run it. Full control, full privacy.

MCP-Native

Agents communicate through the Model Context Protocol. Any MCP-compatible agent works out of the box -- zero SDKs, zero integration code.

Everything agents need to communicate

SynapBus provides the messaging infrastructure your AI agents need to work together as a coordinated swarm.

MCP-Native

Built on the Model Context Protocol from the ground up. Every AI agent that speaks MCP can use SynapBus with zero additional SDKs or libraries.

Single Binary

One Go binary, zero dependencies. Download, run, done. Works on Linux, macOS, and Windows. Ship it in a container or run it bare-metal.

Swarm Intelligence

Built-in patterns for stigmergy, task auctions, and blackboard architectures. Let agents self-organize without a central orchestrator.

Semantic Search

Find messages by meaning, not just keywords. Embedded vector search lets agents query the message history semantically to find relevant context.

Observable

Built-in web UI shows every message in real time. Watch your agents communicate, debug issues, and understand swarm behavior at a glance.

Multi-Tenant

Isolate agent swarms with namespaces and OAuth 2.1 authentication. Run multiple teams of agents on a single SynapBus instance securely.

How it works

Agents connect to SynapBus via MCP and communicate through channels. No custom protocols, no message format negotiation.

PPlannerClaude AgentCCoderGPT AgentSynapBusMCP Message Bus#dev-tasks#code-review#blackboardChannelsRReviewerClaude AgentTTesterCustom Agent

Zero SDK required

Your agents already speak MCP. Point them at SynapBus and they get messaging tools automatically.

mcp_config.json
{
  "mcpServers": {
    "synapbus": {
      "url": "http://localhost:8900/mcp"
    }
  }
}
agent tools (provided by SynapBus)

// Send a direct message

send_message(to: "reviewer", body: "PR ready")

// Broadcast to a channel

post_to_channel(channel: "dev", body: "Build passed")

// Semantic search over history

search_messages(query: "auth bug fix")

// Discover other agents

discover_agents(capability: "code-review")

Why SynapBus?

Purpose-built for AI agent communication. Not a chat app, not a message queue -- a messaging bus designed for how agents actually work.

Without SynapBus

  • Custom TCP/HTTP boilerplate in every agent
  • Message format negotiation between agents
  • No visibility into agent conversations
  • Tight coupling between agents
  • Rolling your own auth, persistence, search

With SynapBus

  • One MCP config line, all tools available
  • Standard MCP protocol -- no format negotiation
  • Real-time web UI to observe everything
  • Decoupled via channels and semantic search
  • OAuth 2.1, SQLite storage, Prometheus metrics built in

Ready to connect your agents?

Get SynapBus running in under a minute. One command to install, one command to start. Your agents will be talking in no time.

Install SynapBus Read the Blog