This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
A small local MCP server for authenticated, explicit-text message exchange between independent clients. It provides a durable SQLite WAL queue, idempotent sends, replay-safe inbox reads, bounded conversations, manual pause/reset controls, and a client-driven polling boundary.
This repository contains generic source only. Participant names, tokens, data paths, and admin identity are configured at runtime. No chat transcript, private memory, database, token, tunnel configuration, provider-specific account setup, or local machine path is included.
Related MCP server: agent-mailbox
python -m venv .venv
.venv\Scripts\python -m pip install -e .
python -m message_bridge.setup_auth --file runtime/auth.json --participants participant-a participant-b --forceThe command prints each token once. Keep tokens local. Do not commit runtime/.
$env:BRIDGE_PARTICIPANT = "participant-a"
$env:BRIDGE_TOKEN = "<local token>"
$env:BRIDGE_AUTH_FILE = "runtime/auth.json"
$env:BRIDGE_DATA_DIR = "runtime"
$env:BRIDGE_ADMIN_PARTICIPANT = "participant-a"
python -m message_bridge.serverStart a separate process for each participant using the same data directory. The MCP tools are:
bridge_send(to, text, message_id, conversation_id)
bridge_inbox(limit, after_id, wait_seconds, acknowledge)
bridge_status(action)
Default limits are 12,000 characters per message, 8 sends per participant per conversation, 16 messages per conversation, and 36 total turns. Change these constants in source for a different deployment policy.
Only the text explicitly passed to bridge_send is queued. The server does not read model chats, private memory, tool output, cookies, or credentials. It cannot wake a model or create a new model turn. bridge_inbox is client-driven polling. acknowledge=false is replay-safe; acknowledge only after the caller has accepted the response.
The server refuses self-send, invalid identifiers, duplicate payload conflicts, paused sends, conversation limits, turn limits, and unauthenticated clients. Logs contain metadata and never message bodies or tokens. Keep the runtime directory private.
MIT. See LICENSE.
This server cannot be deployed
Shared rooms and durable notes for agents over plain HTTP: rendezvous, hand-off, coordination.
Agent-to-agent messaging: directory, public lobby, DMs, channels, search. Stateless MCP + REST.
Public and private rooms for agents, with messages, files, search, and resumable events.
Authenticated event topics for agent-to-agent messaging with per-agent credentials and audit logs.