allstar-mcp

MCP server for AllStar Link node control. Wraps ASL3-API REST endpoints as MCP tools, giving AI agents (Claude, GPT-4, any MCP-compatible client) structured, intent-aware access to AllStar Link node monitoring and control.

The first MCP server for AllStar Link in existence.

Architecture

Claude / AI Agent
      |
  allstar-mcp  (FastMCP, Python)
      |
  ASL3-API  (REST + SSE, port 8073)
      |
  AMI / Asterisk / app_rpt

The MCP server is a translation layer only. It never touches AMI directly — all operations go through ASL3-API REST endpoints.

Requirements

• ASL3-API v1.4+ running on your AllStar node (see ASL3-API)
• Python 3.11+ or uv

Quick Start

ALLSTAR_API_KEY=yourkey ALLSTAR_API_URL=http://your-node:8073 uvx allstar-mcp

Claude Desktop Configuration

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "allstar": {
      "command": "uvx",
      "args": ["allstar-mcp"],
      "env": {
        "ALLSTAR_API_KEY": "your-api-key-here",
        "ALLSTAR_API_URL": "http://your-node-ip:8073"
      }
    }
  }
}

Environment Variables

Tool Surface

Read-only (no risk)

Low-risk control

Medium-risk (require situational awareness)

High-risk (destructive)

Resources

Safety Design

• Active-QSO guard: connect_node and disconnect_node automatically block during active RX/TX unless override_active_qso=True
• Confirmed flag: connect_node, disconnect_node, send_dtmf, execute_macro, and disconnect_all all require confirmed=True and will no-op with an explanation if not set
• Dry-run mode: All confirmed-action tools accept dry_run=True to preview the action and payload without executing
• Audit context: get_audit_log lets the agent check recent command history before issuing duplicate commands
• No AMI access: The server has no Asterisk/AMI credentials and cannot bypass ASL3-API

License

MIT