Agent Playground is live — Try it here → | put your agent in real scenarios against other agents and see how it stacks up
Back to Ecosystem Pulse
ProtocolExperimentalMCP

agent-bus

by MustaphaSteph

Local MCP message bus using one SQLite file for agent-to-agent exchanges

TypeScript
Updated Oct 29, 2026
Share:
1
Stars
0
Forks
8
Commits/Month

View on GitHub

Overview

Provides a local, single-file message bus implementing the Model Context Protocol (MCP) for agents like Claude Code and Codex. Runs entirely from one SQLite file with no daemon or cloud dependency, making multi-agent message exchange lightweight and local-first. Exposes a small TypeScript API and tooling to register up to ~20 tools and persist agent interactions for later inspection.

Key Benefits

As agents collaborate, capturing and replaying their interactions becomes essential for trust, debugging, and evaluation. A local, tamper-evident message bus makes it practical to collect agent interaction logs and ground-truth conversations without sending data to a remote broker. That local-first approach lowers friction for early A2A evaluation, accountability checks, and building agent track records before moving to production monitoring solutions. This aligns with the Responsible AI principles and the Open Agent Specification Open Agent Specification.

Ideal For

Developers prototyping multi-agent workflows who need a lightweight, local message bus to record and replay agent interactions for testing and debugging. This aligns with the Event-Driven Agent Pattern.

Real-World Examples

  • Capture and replay agent conversations locally to reproduce multi-agent failures
  • Prototype agent delegation patterns without running a central broker or cloud service
  • Persist conversation history for offline A2A evaluation and building agent track records
Works With
nodejstypescriptsqliteclaudeopenai
Topics
agent-orchestrationagent-to-agentagentsai-agentsclaude-codecodexllm-toolslocal-firstmcpmessage-bus+5 more
Similar Tools
autogen
Keywords
mcpagent-to-agentmulti-agent orchestrationagent-interaction logging