BMC has added Model Context Protocol capabilities to its mainframe and workflow automation products, allowing AI agents to interact with live operational systems under enterprise controls.
The additions cover BMC AMI Assistant and Control-M, extending access across mainframe, cloud, and hybrid environments.
A new MCP-enabled client for BMC AMI Assistant is designed to give authorised users access to institutional knowledge and current operational information held across different IT teams. Database administrators can also surface live information from BMC AMI Ops within the company’s command environment for Db2.
Control-M gains an MCP server through which authorised agents and assistants can trigger, monitor, and investigate production workflows. BMC says those interactions remain subject to governance, visibility, policy controls, and human oversight.
MCP is an open protocol intended to standardise how AI applications connect with data, tools, and services. Adoption can reduce the need to construct a separate integration for every model and enterprise platform.
An assistant connected only to documentation may explain how a process works. Once connected to a production workflow, an agent can potentially inspect current conditions, retrieve operational information, and initiate an action.
BMC has also expanded the Control-M Archive Service across self-hosted and software as a service environments. The service adds cloud archiving, centralised change management integrated with IT service management systems, and workflow histories intended to support audits and troubleshooting.
New Control-M integrations include AWS RDS, Oracle Data Transform, SAP CPI, Azure VM Scale Sets, Azure AI Foundry, and Dataiku. The platform can therefore coordinate conventional applications, data pipelines, AI agents, and AI-assisted tasks through the same operational environment.
Further mainframe additions include software bills of materials within BMC AMI DevX Code Pipeline, allowing teams to identify common vulnerabilities and the applications affected by them during continuous integration and deployment.
BMC AMI Ops Monitoring gains context-aware analytical alarms intended to distinguish material changes in system behaviour from ordinary operational noise. The platform combines visibility across z/OS systems and containerised zCX workloads.
Ram Chakravarti, chief technology officer at BMC, said: “AI creates enterprise value only when it can understand operational context and take action within the guardrails the business requires. By bringing MCP capabilities to BMC AMI Assistant and Control-M, we are enabling AI agents to work securely with live operational data and production workflows while preserving governance, visibility, and human control. When organisations come to BMC first, their outcomes are reliable, governed, explainable, and in sync, so digital business can move faster without losing control.”
Allowing AI systems to execute multi-stage work creates greater operational exposure than using them to generate summaries or answer questions. Identity errors, weak instructions, unauthorised access, and incorrect decisions can have consequences across payroll, logistics, billing, settlement, customer service, and production.
Research into autonomous agents has already exposed substantial governance weaknesses, including organisations reversing automated actions and postponing deployment because failures could not be traced adequately.
Mission-critical systems demand a higher standard of control than general productivity software. A poor answer in a chat window may waste an employee’s time; a poor action in a production environment can interrupt services, alter records, or create financial and regulatory consequences.
MCP may simplify connectivity, but standardisation alone does not provide security. Organisations must still determine the identity of each agent, the systems it can access, the data it can retrieve, and the actions it is allowed to perform.
Least-privilege access will be central to safe deployment. An agent permitted to investigate a failed workflow does not automatically need authority to restart it, alter a schedule, or change production data. Permissions should reflect the narrowest action required by the approved use case.
Human approval also needs careful design. Requiring an employee to approve every low-risk step can eliminate much of the efficiency, while allowing unsupervised execution across high-impact systems creates unacceptable exposure. Risk tiers can determine which actions are automatic, supervised, or prohibited.
Audit requirements become more demanding when several models, tools, and data sources contribute to a decision. Logs may need to record the user request, model, prompt, retrieved information, tools invoked, permissions applied, output generated, approval obtained, and final action.
Mainframes create additional institutional complexity. Many organisations rely on them for high-volume transactions and critical records while facing a shortage of experienced specialists. AI assistants may help employees retrieve knowledge and interpret operational information, but teams must retain the capacity to verify outputs independently.
ServiceNow has also expanded controls across models, agents, identities, and external systems, indicating that enterprise technology suppliers are competing on governance, orchestration, and accountability as well as model capability.
As several agents and tools become connected across mixed technology estates, interoperability, observability, and policy enforcement will carry greater weight in purchasing decisions.
BMC’s additions place agents alongside systems that organisations often describe as unable to fail. Adoption will depend on whether customers can secure faster investigation and automation without weakening change control, segregation of duties, resilience, and accountability.
You must be logged in to post a comment.