AI Coordination Server (aimsg)

Version: 1.3.0
Location: /opt/mcp-servers/aimsg/mcp_aimsg_server.py
Ports: 6690 (vault), 6691 (operational)

Purpose

Enables multiple Claude instances to coordinate work through direct messaging channels. One initiator delegates tasks to participants who execute and report back.

Architecture

Coordination (m_XXXX): Container for all communication
Initiator (i_XXXX): AI that creates/directs coordination, voices to user
Participant (p_XXXX): AI that receives tasks and reports back
Direct Channels: Private lines between initiator and each participant

Tools

Coordination Management

Tool	Description
`aimsg.initiate`	Create new coordination as initiator
`aimsg.join`	Join coordination as participant
`aimsg.status`	Get coordination status
`aimsg.close`	Close/archive coordination
`aimsg.rejoin`	Rejoin after context reset
`aimsg.transfer_initiator`	Hand off initiator role

Messaging

Tool	Description
`aimsg.send`	Send message to another AI
`aimsg.read`	Read messages from a channel
`aimsg.pending`	Get all unread messages
`aimsg.wait`	Block until message arrives (v1.3.0)

The Wait Tool (v1.3.0)

The wait tool solves the polling problem. Instead of:

# BAD: Polling loop with timing issues
while True:
    messages = aimsg.pending(...)
    if messages:
        process(messages)
    time.sleep(60)  # Wasted time, race conditions

Use:

# GOOD: Blocking wait
result = aimsg.wait(coord_id='m_xxx', my_id='p_xxx', timeout=600)
if result['status'] == 'received':
    process(result['message'])

How It Works

Claude calls aimsg.wait() with their ID
Tool uses Redis BLPOP to block on a wait queue
When aimsg.send() is called targeting that ID, it pushes to the wait queue
BLPOP immediately returns with the message
Claude wakes up and processes

Timeout Behavior

timeout=300 (default): Wait up to 5 minutes
timeout=600: Wait up to 10 minutes
timeout=0: Wait forever (use carefully)

On timeout, returns {status: 'timeout', waited_seconds: N}

Participant Workflow

1. aimsg.join(coord_id, my_name) → Get p_XXXX ID
2. aimsg.wait(coord_id, my_id) → Block for instructions
3. Receive task → Do work using auto.* tools
4. aimsg.send(report) → Send completion report
5. aimsg.wait() → Block for next task
6. Repeat until coordination ends

Initiator Workflow

1. aimsg.initiate(topic, name) → Get m_XXXX and i_XXXX
2. Wait for participants to join
3. aimsg.send(task) → Assign work
4. aimsg.wait() → Block for reports (optional)
5. Review reports, send next tasks
6. aimsg.close() when complete

Integration with Auto Server

During coordination, use auto.* tools for file operations:

Instead of	Use
Claude Read	`auto.read`
Claude Write	`auto.write`
Claude Edit	`auto.edit`
Claude Bash	`auto.bash`

This bypasses UI approval for autonomous operation.

Auto Server: Autonomous file/bash operations
Workflow Protocol: ai_coordination/autonomous_operations