Skip to main content
Glama
Agent Communication Bridge

Agent Communication Bridge

by Ezhil948

agent-communication-bridge

Local Model Context Protocol (MCP) server, registry, and client hooks for messaging between Antigravity (agy), OpenAI Codex (codex), and OpenCode (opencode) sessions on the same machine.

How it works

  • Registry (http://localhost:4999): An Express + WebSocket + SQLite server that tracks active sessions, stores messages, and records delivery states (queued, delivered_to_bridge, displayed_to_client, answered, failed).

  • MCP server (acb mcp start): Exposes messaging tools inside each client and resolves the active workspace path from the host or parent process so separate projects do not collide on a plugin directory.

  • Session isolation: Each live session gets a unique agentId (client-<type>-<projectHash>-<sessionTag>). Multiple windows or sessions in the same project stay separate without kicking each other off the WebSocket.

  • Turn delivery: Antigravity hooks (SessionStart / Stop + Language Server), Codex app-server (codex >= 0.162 or acb codex app-bridge), and the OpenCode plugin start a turn in the receiving session when a message arrives.

Related MCP server: hauddy

Requirements

  • Node.js >= 22

  • pnpm

  • At least one client installed: agy, codex, or opencode

Install

git clone https://github.com/Ezhil948/agent-communication-bridge.git
cd agent-communication-bridge
pnpm install
pnpm run build
npm install -g .

On macOS, Linux, or Git Bash you can run bash bin/install.sh instead.

This installs both agent-communication-bridge and acb on PATH.

Setup

Start the registry and configure the clients you use (one-time per machine):

acb up
acb antigravity install-plugin
acb init --client codex --yes
acb opencode install-plugin --global
acb doctor

For per-project configuration (writes .mcp.json and optionally scopes sessions with --identity), run acb init inside the project directory.

Usage

  1. Open your project in two clients (or start a headless Codex worker with acb codex app-bridge --project "<path>").

  2. From one session, call list_agents(includeClients=true) to find the target's agentId.

  3. Call message_client_session(clientId, message, expectsResponse=true). Optionally pass model or reasoning to override them for that turn.

  4. The receiving session wakes up, reads channel_inbox(pendingOnly=true) if needed, and answers via reply using replyWith.agentId, replyWith.conversationId, and replyWith.replyTo.

MCP tools

Tool

Purpose

list_agents

List connected agents and client sessions (includeClients=true).

message_client_session

Send a message by clientId or project, with optional expectsResponse, model, and reasoning.

channel_inbox

Read pending or recent threads (pendingOnly=true / false).

reply

Reply to a thread using the replyWith fields from channel_inbox.

channel_clear

Clear answered, failed, all, or a specific conversationId.

agent_bridge_guide

Return inline usage docs by topic (overview, setup, send, reply, acks, troubleshooting).

CLI

acb up                         # Start background registry on :4999
acb down                       # Stop background registry
acb status                     # Show registry status and agent count
acb list                       # List registered sessions
acb prune                      # Remove stale entries
acb doctor                     # Check Node, PATH, registry, and client plugins
acb init [--project <path>]    # Write .mcp.json and configure clients
acb antigravity <cmd>          # install-plugin | attach | hook-stop | hook-session-start
acb codex <cmd>                # start | app-bridge | tmux-bind | tmux-sidecar
acb opencode <cmd>             # start | bridge | install-plugin
acb prompt <cmd>               # path | show | init | edit | reset

Notes

  • Separate windows, CLI processes, and projects run as isolated sessions. If one IDE window shares a single MCP child process across multiple tabs, those tabs share that client connection unless separated with --identity.

  • See docs/architecture.md, docs/cli-reference.md, and CONTRIBUTING.md for details.

License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Enables multiple coding agents (Claude Code, Codex, Cursor) to discover each other's sessions, search transcripts, ask questions, and handoff tasks through a shared MCP server.
      5
      6 npm
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI agents to message each other by @nickname via an MCP server, with contacts, presence, and durable delivery across local and remote agents.
      3
      Apache 2.0
    • A
      license
      B
      quality
      B
      maintenance
      Enables AI coding agent sessions working in the same project folder to discover one another, message each other, track status, and bid on owner-posted tasks over MCP. Also lets a human watch the conversation live on a chat page and reply to one agent or all of them.
      14
      1
      AGPL 3.0
    • A
      license
      B
      quality
      B
      maintenance
      Enables a lead AI agent to spawn and coordinate manager agents, which can in turn run child agents across Codex/ChatGPT, Google Antigravity, Claude Code, and local OpenAI-compatible models. Provides messaging, waiting, transcripts, file exchange, and saved-conversation adoption so multi-agent workflows can be orchestrated from MCP hosts.
      14
      MIT