Agent Orchestrator MCP Server

By MEOK AI Labs — Sovereign AI tools for everyone.

Agent Orchestrator MCP Server

Multi-agent task management system for AI applications. Create agents with roles and capabilities, delegate tasks with trust-based routing, coordinate file access to prevent conflicts, run focused sprints, and monitor performance through a unified dashboard.

Based on the Sovereign Temple 47-agent coordination framework, simplified for standalone use. Data persists in ~/.mcp-agents/.

Tools

Installation

pip install mcp

Usage

Run the server

python server.py

Claude Desktop config

{
  "mcpServers": {
    "agent-orchestrator": {
      "command": "python",
      "args": ["/path/to/agent-orchestrator-mcp/server.py"]
    }
  }
}

Example workflow

1. Create agents:

Tool: create_agent
Input: {"name": "Research Bot", "role": "researcher", "department": "research", "capabilities": ["web_search", "analysis"]}
Output: {"status": "created", "agent_id": "research_bot", "role": "researcher"}

2. Delegate a task:

Tool: delegate_task
Input: {"task": "Research competitor pricing models", "capability": "web_search", "priority": "high"}
Output: {"status": "delegated", "task_id": "a1b2c3d4", "agent_id": "research_bot"}

3. Coordinate file access:

Tool: acquire_files
Input: {"agent_id": "research_bot", "files": ["report.md", "data.json"], "task_id": "a1b2c3d4", "exclusive": true}
Output: {"status": "acquired", "files": ["report.md", "data.json"]}

4. Complete the task:

Tool: complete_task
Input: {"task_id": "a1b2c3d4", "agent_id": "research_bot", "result_summary": "Found 5 competitor pricing tiers...", "care_score": 0.8}
Output: {"status": "completed", "task_id": "a1b2c3d4"}

5. Check the dashboard:

Tool: get_dashboard
Output: {"agents": {"total": 3, "active": 3, "avg_trust": 0.52}, "tasks": {"total": 12, "by_status": {"completed": 8, "assigned": 4}}, ...}

Trust System

Agents accumulate trust through successful task completion:

• Successful task: trust += 0.02 x care_score (max 1.0)
• Failed task: trust -= 0.05 (min 0.0)
• Auto-routing prefers higher-trust agents
• Trust persists across sessions

Data Storage

All data persists in ~/.mcp-agents/:

• agents.json - Agent registry
• tasks.json - Task history
• sprints.json - Sprint records

Pricing

License

MIT