Back to Projects
Repowire - Mesh network for AI coding agents - enables Claude Code sessions to communicate with each other in real-time

Repowire

January 24, 2026
Featured
python AI Claude Code MCP developer tools agents open source
View on GitHub View Project

The Problem

Coding agents like Claude Code are great at understanding repositories, but they work in isolation. One repo, one session. When you’re working across multiple repos—frontend and backend, app and infrastructure, microservices—you become the message relay between agents that can’t talk to each other.

Repowire fixes that.

What It Does

Repowire creates a mesh network where Claude Code sessions can communicate directly. The agent in your frontend repo can ask the one in your backend repo a question and get a real answer based on the actual code, not stale documentation.

"Ask backend what API endpoints they expose"

Claude uses an ask_peer tool, the query routes to the backend session, and the answer comes back. No human relay. No copy-paste. Live context from the source of truth.

How It Works

Three components work together:

  • Daemon: Routes messages between peers, maintains registry of active sessions
  • Hooks: Integrate with Claude Code to register sessions and capture responses
  • MCP Server: Provides ask_peer, list_peers, notify_peer, broadcast tools

Sessions auto-discover each other. Each remains specialized in its own repo while being able to reach out to others when needed.

Quick Start

# Install
uv tool install "repowire[claudemux]"

# Setup (installs hooks, MCP server, daemon)
repowire setup
repowire status

# Open Claude in two repos (in tmux)
tmux new-session -s dev -n frontend
cd ~/projects/frontend && claude

# In another window
tmux new-window -t dev -n backend
cd ~/projects/backend && claude

Now frontend-Claude can ask backend-Claude questions directly.

Features

  • Pull-based context: Agents request context when they need it, from the source of truth
  • Auto-discovery: Sessions register automatically via hooks
  • Real-time communication: Sync queries and responses, not stale docs
  • Dashboard: Monitor peer status and communication at localhost:8377/dashboard
  • Complementary: Works alongside memory banks and documentation

Read More

See the full story in the blog series: