Omnissa Elara AI Governance, Cloud PC and MCP Server: What Windows Admins Can Use Now

The announcement at a glance​

Omnissa made the announcement on September 29, 2026. It said the new products would expand AI-ready workspaces for people and AI, bring greater autonomy to IT operations, and introduce new governance capabilities over applications, agents and high-impact actions.

The company's pitch rests on its own research. It says its State of the Digital Workspace report found that AI assistant usage across enterprise endpoints increased nearly 1,000% year over year, with three-quarters of it coming from unsanctioned tools. The separate Elara press release says Omnissa research found that unsanctioned AI tools were present on 75% of enterprise-managed devices. These are vendor figures, and the two may not measure the same thing. Treat them as marketing context rather than independent findings.

Summary: This is a set of products, not one launch, and each piece has a different release status. The status table further down sorts out what's real now.

Elara: an "authority layer" for shadow AI and risky changes​

Omnissa calls Elara a new authority layer for AI governance and high-impact enterprise actions. In practice it sits above your existing tools. It pulls signals from identity, device, app and AI systems and checks them against your policies before something consequential happens. The company says it is designed to sit above these tools, connecting the systems that customers already run without forcing you onto a single technology stack.

The beta covers four areas:

• Shadow AI discovery. Finding AI apps, LLMs, agents and other tools, including ones outside approved inventories. The product page says discovery spans AI apps, models, agents, MCP servers, and tools in use across managed endpoints. The word "managed" matters: unmanaged BYOD devices appear to be outside that scope.
• AI usage guardrails through the Omnissa AI Gateway. Omnissa says Elara can authorize model access, apply usage guardrails and meter token consumption across providers.
• Change control for high-impact actions. Elara checks proposed changes against conflicts, freezes and dependencies. Based on that context, it either clears the action, bounds it, or routes it to the approver who holds the authority.
• Audit evidence. It records what happened, who authorized it and what it touched as a replayable record.

One detail is easy to miss. Elara does not block everything itself. According to Omnissa, where Elara holds an enforcement point it intervenes directly; elsewhere it advises, routes, or remediates. If none of your existing systems gives Elara an enforcement point, it works more like a well-informed advisor than a firewall.

On privacy, Omnissa says Elara works from operational signals such as identities, device posture, events and approvals, rather than the content of an employee's work. It also says data handling and retention detail will be published alongside the datasheet. Until that detail is published, compliance teams have a reasonable question to put to the vendor.

Status: Elara is now available in beta, and the product page still asks people to join a waitlist. It is not generally available.

Cloud PC and Horizon Delegate​

Omnissa Cloud PC combines compute, Workspace ONE UEM, digital employee experience (DEX) monitoring and Horizon Cloud DaaS into one service run by Omnissa. It will initially run on AWS, with additional hyperscaler options planned. Omnissa's launch blog adds that it is targeted for availability early next year, which means early 2027. A bring-your-own-capacity option on AWS is also planned.

Organizations already on Workspace ONE get one practical benefit. Cloud PC is managed only through Workspace ONE UEM, and existing tenants can be reused. That puts it in direct competition with Windows 365's turnkey pitch. The difference is the management console: Omnissa is selling to shops that run Workspace ONE rather than Intune.

Horizon Delegate is an agent that runs inside a user's Horizon VM and does tasks on the user's behalf. The notable part is that it can continue working even when the user is not actively connected to the virtual session, with the same entitlements and controls as the user. TechTarget's coverage of the show says Delegate is expected in limited availability with Horizon 2609. Omnissa's own release gives no general availability date and no list of supported tasks.

So an agent that keeps working after you close the laptop, using your identity, is the feature. For security teams it's also the reason to read the fine print before turning it on.

Three agents for IT, one built for Windows​

Omnissa announced three agents for IT operations:

1. DEX agent. It helps investigate an employee experience issue, identify root cause and use existing playbooks or prior Workspace ONE Assist session history to recommend a resolution. It recommends a fix; it doesn't apply one.
2. Workspace ONE Vulnerability Defense agent. It reviews and prioritizes vulnerabilities and prepares patching actions, with human-in-the-loop guardrails. Omnissa justifies the speed by noting that time to exploit vulnerabilities now less than a day. Don't read this as an agent that patches everything on its own: the human approval step is part of the design.
3. Windows App Lifecycle agent. Omnissa acknowledges that packaging Windows apps often requires significant manual work from IT teams. The agent packages and tests an app, then hands the result to an administrator for review.

Anyone who has fought a stubborn MSI transform will see the appeal of the third one. Omnissa has not published supported Windows builds, app catalogs, licensing or rollout dates for any of these agents, so test before you trust it.

The hosted MCP Server works more like an API key than a chatbot​

The hosted Omnissa MCP Server lets admins connect the AI tool of their choice, such as Claude or another MCP client, to the Omnissa platform. Omnissa says the services available to the agent are based on the administrator's existing role and access permissions.

Omnissa's community technical blog gives more detail on how it works. You'll need your organization onboarded to Omnissa Connect first. From there, an Org Owner or Administrator turns on MCP access for other admins. Each admin gets either Read access (view only) or Full access (read, modify, execute), and that scope can never go beyond what their existing product role already allows. Every token is logged with the user and scope attached.

Basic setup, as described in Omnissa's beta documentation:

1. Make sure your org is onboarded to Omnissa Connect and that an Org Owner or Administrator has enabled MCP access for you.
2. Get your token and URL from Omnissa Connect under Account settings > MCP Tokens. Update your client config to point at the Omnissa MCP Server, passing your token as a bearer header through mcp-remote.
3. Start with a Read-scoped token. Move to Full scope only after you've validated workflows.

Omnissa's own warning is the most useful line in the documentation: since an MCP token carries your admin scope into whatever AI client holds it, we recommend testing workflows in UAT first and keeping a person in the loop for anything that writes data. In other words, treat a Full-scope MCP token like a domain admin password you've handed to a chatbot. Other limitations in the beta: OAuth isn't supported yet in this beta, and the initial beta covers UEM, Access, Intelligence, and Hub Services. Horizon Cloud support is coming in a future release.

The coverage is inconsistent on Horizon. TechTarget's write-up says the server spans Workspace ONE, Horizon, DEX and Intelligence, while Omnissa's early-September beta posts list Horizon as coming later. If Horizon tooling is what you need, check what your tenant actually exposes rather than relying on either account.

Coverage from VMblog of a live demo shows why scoping matters. In the demo, an admin's AI client found a Horizon pool sizing problem but couldn't change anything, because Jack's token for Horizon was read-only, so it offered to open a change ticket instead. That is least privilege doing its job.

Status check: what you can actually use​

The MSP angle, and some skepticism​

Channel Insider's coverage leans heavily on Omnissa CMO Renu Upadhyay. She argues that "MSPs now have the chance to move from managing employees' devices to running secure environments where people and AI agents work." She also says customers "want fewer disruptions, better employee experiences, and stronger security" rather than ticket counts.

That's a fair point about where managed services are heading, but it comes from the vendor selling the tools. No customer data yet shows that these agents cut headcount needs or patch backlogs. The independent analysts quoted in Omnissa's materials, IDC's Phil Hochmuth and Omdia's Adam Holtby, endorse the problem Omnissa describes rather than proven results. Holtby puts the real challenge as connecting existing control points into a governance model that can keep pace as AI becomes more autonomous.

What Windows admins should do now​

• Run your own shadow AI inventory. Elara is still in beta, so use existing UEM, EDR and network telemetry to find AI apps and local MCP servers on managed endpoints. You need that baseline whichever vendor you pick.
• Treat MCP tokens as privileged credentials. Use Read scope by default, keep token lifetimes short, revoke tokens nobody can account for, and test in UAT.
• Pilot the Windows App Lifecycle agent on low-risk apps once it reaches your tenant, and keep administrator review in place.
• Ask about data retention before you connect Elara to your identity and ITSM systems.
• Compare Cloud PC against Windows 365 on management fit. A Workspace ONE shop and an Intune shop will reach different answers.

Omnissa's direction is a sensible one: govern AI actions before they happen instead of cleaning up afterward. For now, though, most of this announcement is beta, roadmap or target date, and the strongest evidence for it comes from the vendor.