AI agents

Use @palmpush/mcp when an MCP client or AI agent should send notifications to you.

The server exposes one tool:

txt

send_push_notification

Single listener

Use this when every agent notification should go to one listener.

bash

PALMPUSH_LISTENER_ID=abc123abc123abc123abc123abc123ab npx @palmpush/mcp

The tool input is simple:

json

{
  "title": "Approval needed",
  "body": "The agent is waiting for your decision."
}

Named listeners

Use a private config file when the agent should choose between multiple listeners.

json

{
  "listeners": [
    {
      "name": "approval",
      "listenerId": "aaa123aaa123aaa123aaa123aaa123aa",
      "description": "Use when the user needs to decide."
    },
    {
      "name": "status",
      "listenerId": "bbb123bbb123bbb123bbb123bbb123bb",
      "description": "Use for non-blocking status updates."
    }
  ]
}

Start the server with:

bash

PALMPUSH_CONFIG=/path/to/palmpush-mcp.json npx @palmpush/mcp

In this mode, the tool also requires listenerName:

json

{
  "title": "Task finished",
  "body": "The report is ready.",
  "listenerName": "status"
}

MCP client config

json

{
  "mcpServers": {
    "palmpush": {
      "command": "npx",
      "args": ["-y", "@palmpush/mcp"],
      "env": {
        "PALMPUSH_CONFIG": "/path/to/palmpush-mcp.json"
      }
    }
  }
}

Security

Listener ids are secrets. Keep MCP config files private and avoid placing them inside project folders that AI agents can freely read.