IBM MQ Agent 2.1.0 adds A2A links and Red Hat AI inference choices

IBM has released MQ Agent 2.1.0, adding Agent-to-Agent protocol support and widening the inference backends that can power the messaging specialist. The release is documented in IBM’s MQ Agent change log, which IBM marked as updated on October 21.

The Red Hat connection is direct: IBM’s engineering post for the release lists managed Red Hat AI Inference alongside watsonx.ai and Amazon Bedrock. For self-managed inference, it lists OpenShift AI, RHEL AI and upstream vLLM.

MQ expertise becomes callable

Before this release, customers mainly reached the MQ Agent through the MQ Console. IBM says A2A support now lets compatible agents and orchestration systems call it as a specialist inside a wider workflow, such as asking whether queues are building up or a channel has stopped during an incident investigation (IBM Community).

That matters because it separates messaging expertise from the coordinating agent. A platform or observability agent can delegate the MQ portion of an investigation rather than reproducing that domain knowledge itself, then combine the answer with evidence from the application and Kubernetes layers (IBM Community).

IBM also describes a scale-out pattern for larger estates. One MQ Agent can cover as many as 20 queue managers; a coordinating agent can route questions to several separately scoped MQ Agents and combine their results. The per-agent limit remains in place, and IBM recommends retaining boundaries between regions, business units, security zones or production stages (IBM Community).

A Red Hat deployment path

The documented architecture runs the MQ Agent on Red Hat OpenShift or Amazon EKS. It combines a supervisor, specialist agents, an MQ documentation specialist and an MCP server that provides read-only access to MQ information (IBM Community).

For OpenShift users, version 2.1.0 therefore creates two distinct Red Hat paths: consume managed Red Hat AI Inference, or keep inference self-managed through OpenShift AI, RHEL AI or vLLM. IBM lists openai/gpt-oss-120b as the primary reasoning model and ibm/granite-4-h-small as an optional secondary model in the published architecture (IBM Community).

The release also tightens generated MQSC commands. IBM’s product documentation says the agent now validates command names, parameters and values, and customizes commands when a queue manager is specified. It does not verify that a valid command will succeed under the caller’s actual authorities.

Teams evaluating the update should test A2A routing and access boundaries together: the useful new capability is broader orchestration, but IBM’s design still depends on each MQ Agent retaining a clearly controlled scope (IBM Community).