Tools Overview

KubeMQ exposes its messaging operations to AI models as MCP tools. A client discovers them with tools/list and invokes them with tools/call — there is no KubeMQ-specific client library involved, just the Model Context Protocol.

Overview

The MCP connector publishes 15 tools in two families:

• 11 core messaging tools — always available. They cover queues, events, the events store, commands/queries, and channel discovery.
• 4 agent-bridge tools — available only when the A2A agent registry is present. They let an MCP client discover and message registered agents, bridging MCP to the A2A gateway.

Every tool maps onto a single KubeMQ messaging operation. The model calls the tool by name with a JSON arguments object; the connector translates it into a native KubeMQ call over the Array and returns the result.

How it works

The connector advertises each tool through tools/list, then routes each tools/call to the matching KubeMQ operation. Core tools reach the broker directly; bridge tools forward through the agent registry to an external agent.

One MCP connector fans out to core messaging tools and, when the registry is present, agent-bridge tools.

Core messaging tools

These 11 tools are registered unconditionally — start kubemq-server and they are live at /mcp.

Agent-bridge tools

These 4 tools appear in tools/list only when the A2A agent registry is injected into the MCP connector. They turn an MCP client into an A2A caller: agent_send builds a message/send envelope and forwards it over Query to _AGENTS_.agents/<agent_id>, while agent_query forwards an arbitrary JSON-RPC method to the agent.

Calling a tool

Every tool is invoked the same way — a tools/call JSON-RPC request naming the tool and passing its arguments object. The skeleton below works for any of the 15 tools; only name and arguments change.

curl -X POST http://localhost:9090/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json' \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "queue_send",
      "arguments": {
        "channel": "example-queue",
        "body": "Hello from MCP"
      }
    }
  }'

Response shape

tools/call always returns a result with a content array. Each entry is a typed block — KubeMQ uses text blocks carrying the operation result as a JSON string.

A successful call:

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [{ "type": "text", "text": "Message sent successfully to queue 'example-queue'" }],
    "isError": false
  }
}

A failed call sets isError: true and carries the error message in the same content block — the JSON-RPC envelope itself still succeeds:

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [{ "type": "text", "text": "error: channel name uses reserved prefix '_AGENTS_.'" }],
    "isError": true
  }
}

Related