Red Hat ships MCP gateway and agent controls in OpenShift 5

Red Hat announced the general availability of its agentic AI management tools for OpenShift 5 at this week's KubeCon conference in Salt Lake City, including an MCP server and an MCP gateway, with MCP support for RHEL and Developer Hub following soon.

The release enters Red Hat into what analysts describe as a race to become the central control plane for enterprise AI agents. Managing autonomous agents has become a top concern for enterprises as AI models grow smarter and more capable, because without adequate controls those agents can cause significant harm. Every major enterprise platform provider is now shipping agentic capabilities: companies can build agents with AWS, Azure, and Google Cloud, with Salesforce, SAP, and Databricks, and with OpenAI and Anthropic, among dozens of other platforms. Many companies use multiple agent builders at once, which makes coordinating all of them a significant challenge.

Choosing a central control plane is one of the biggest decisions companies face today, and Red Hat is now one of the options, on the strength of its MCP support, guardrails, and other agent management features.

Why the 'air traffic controller' race matters

"Everyone wants to be the air traffic controller for AI agents," says Mauricio Sanchez, an analyst at the Dell'Oro Group. "They all want to be that secure infrastructure software layer that will help enterprises as they deploy agents, do so securely."

Red Hat's advantage is portability. Its stack runs across cloud and on-premises environments and works with AI systems from different hyperscalers, AI companies, and enterprise software providers, which helps enterprises avoid vendor lock-in. The trade-off, experts say, is that its capabilities lag behind some competitors, and there are limits to how much oversight it can offer into agents hosted on other platforms.

"An enterprise that has a significant amount of investment in server and software infrastructure should definitely shortlist someone like Red Hat," Sanchez says. "Those enterprises that will continue to host their own server, storage, and compute infrastructure will need software to run that, and the Anthropics and the hyperscalers aren't going to help them do that in the agentic AI era."

Whether Red Hat can expand its share of agentic AI management remains an open question, because the field is evolving quickly. "The race hasn't been won," Sanchez says. "It's the first inning of the race, and it will be fascinating to see where the industry goes."

The commercial opening exists because agentic AI systems are typically built in an API-driven world. AI agents connect to data, tools, and other systems through MCP servers — the Model Context Protocol, an open standard — among other open standards.

"There are still pockets where you can't do everything through an API," Sanchez says. "But if there's one thing that the AI community has embraced is that everything has to be API driven, and if there are pieces that can't be controlled, it's really a question of when those APIs are going to be added."

Who is Red Hat up against?

Red Hat faces substantial competition. Microsoft, in particular, is very strong in this space with Azure Agent 365, says Donovan Brady, vice president of developer relations at Leapwork, a software testing company.

"It has agent support for basically any agent wherever you deploy it," Brady says of Microsoft's offering. "Red Hat definitely does not have that same single control plane."

That gap in reach is the central question hanging over Red Hat's agentic push: enterprises running heterogeneous agent fleets may need a control plane that spans platforms Red Hat cannot fully see into.

What does the MCP tooling actually do?

According to Red Hat's OpenShift 5 announcement, the new MCP gateway sits between AI agents and the tools those agents call, filtering what each agent can see based on its credentials. The release includes NeMo Guardrails, an open source toolkit from Nvidia that can block prompt injections (malicious inputs designed to hijack an agent's behavior) and protect sensitive data. The gateway also lets companies require human approval for any agent action that changes systems.

The new MCP server for OpenShift ties agent access to existing identity systems and Kubernetes role-based permissions, and produces audit trails for agent activity. An agent sandbox is also available in technology preview; it can isolate untrusted agent code in virtual machines.

"Nobody is going to hand an autonomous agent the keys to production on day one. Red Hat seems to understand that," says Matt Kimball, an analyst at Moor Insights & Strategy. "Its agentic story isn't about agents acting on their own. It's about giving them governed access, keeping a human in the loop on remediation, and being able to see what an agent did and what it cost."

Red Hat also announced an MCP server for RHEL (Red Hat Enterprise Linux) arriving in the coming weeks. Its goal is to let enterprises use AI agents to diagnose RHEL issues without granting them unrestricted access.

A further MCP server is coming in Red Hat Developer Hub 2.1, expected to be generally available later this month. That product is an enterprise developer portal built on the open-source Backstage project, and it gives developers, AI development tools, and AI agents access to enterprise context — including a company's software catalog, live operational data, documentation, and self-service workflows.

"Red Hat Developer Hub 2.1 provides a governed foundation for connecting developers and AI agents with that context, helping organizations accelerate software development without sacrificing security, consistency or control," says Balaji Sivasubramanian, Red Hat's head of product for developer tools.

With MCP, Red Hat says, AI agents can access enterprise systems and take actions such as triggering a golden path (a pre-approved development workflow) or debugging a Kubernetes deployment issue.

"MCP is table stakes for any modern platform," says IDC analyst Jim Mercer.

The significance of the announcements, Mercer says, is that the MCP support extends agentic capabilities beyond OpenShift itself. "It seems like Red Hat is positioning OpenShift 5 as an agentic operations platform, not just an infrastructure platform," he says.

OpenShift 5 is the first major release in about seven years; OpenShift 4 shipped in 2019. With this release, Mercer says, "OpenShift is competing not only in Kubernetes and hybrid cloud infrastructure, but increasingly in virtualization, AI infrastructure, agentic operations, and enterprise security."

What else shipped: assistant upgrades, quantum-safe crypto, free virtualization

Beyond agent management, Red Hat promoted OpenShift LightSpeed from a conversational assistant chatbot to a full-fledged AI agent that can diagnose cluster problems, propose fixes, and assess the risk of updates. In a technology preview, LightSpeed can also take on more troubleshooting tasks, with a human still in the loop to approve every change.

AI was not the only focus this week. OpenShift 5 also adds hardened multi-tenancy for sovereign deployments and expands support for post-quantum cryptography (encryption designed to resist attacks by future quantum computers).

Red Hat is additionally offering a free year of OpenShift Virtualization for qualifying three-year agreements — a move that could attract VMware customers looking to leave that platform without paying twice during the transition period.

Analysts are skeptical about the migration incentive. OpenShift Virtualization is "technically viable" as a VMware alternative for on-premises deployments, says Leapwork's Brady. "But it's a very different platform and you would need to completely rearchitect the way your data center is built and operated," he says. "You're talking about a significant amount of time and effort and money. It might not be worth it."

Dimitris Mavrakis, senior research director at ABI Research, agrees that switching a virtualization technology stack is a significant effort. "We don't believe the one-year free license argument to be the strongest catalyst for migrating," he says.

With MCP servers landing across RHEL and Developer Hub within weeks, Red Hat has committed to a specific bet: that enterprises will want their agent control plane anchored in the same infrastructure software that already runs their data centers. Whether that bet beats Microsoft's broader single control plane is, in Sanchez's framing, a race still in its first inning.