Claude Bridge MCP

MCP server enabling communication between multiple Claude Code instances via a shared message bridge.

Features

• bridge-join — Register as an agent on the bridge
• bridge-send — Send a message to another agent or broadcast to all
• bridge-read — Read unread messages (marks them as read)
• bridge-agents — List all registered agents
• bridge-history — Get the full conversation history
• bridge-reset — Clear all messages and agents

How it works

Multiple Claude Code terminals connect to the same bridge (a shared JSON file). Each instance registers with a name and project, then communicates via messages. A coordinator can orchestrate the work.

Terminal 1 (erezo-dev)  ←→  bridge.json  ←→  Terminal 2 (newsletter-dev)
                               ↑
                      Terminal 3 (coordinator)

Installation

npm install
npm run build

Configuration

Add to your Claude Code MCP config (~/.claude.json):

{
  "mcpServers": {
    "claude-bridge": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/claude-bridge-mcp/dist/index.js"]
    }
  }
}

Environment variables

| Variable | Default | Description | |----------|---------|-------------| | BRIDGE_PATH | /tmp | Directory where the bridge JSON file is stored |

Usage

In each Claude Code terminal:

1. Join the bridge: bridge-join with a name and project
2. Send messages: bridge-send to a specific agent or "all"
3. Read messages: bridge-read to check for new messages
4. Monitor: bridge-history to see the full conversation

Use cases

• Multi-project coordination: Two devs working on interdependent projects (e.g., API + frontend)
• Project management: A coordinator assigns tasks and tracks progress
• Code review: One agent reviews another's work via messages
• Pair debugging: Two agents collaborate on a bug across repos

License

MIT