Short version: this is a read-only data pipe that an admin has to turn on. It is not a robot that patches your fleet while you sleep.
What Ivanti actually shipped
Ivanti announced the capabilities on November 18, 2026. Its release page labels MCP Server access as GA. Per that page, customers entitled to the Ivanti Neurons Platform can now enable read-only MCP Server access. Activation is admin-controlled. Nothing is switched on automatically for every tenant.
MCP is an open standard created by Anthropic. It gives AI systems a consistent way to connect to tools, data sources, resources and prompts. In Ivanti's framing, Neurons stays the system of record and MCP is only the access protocol. Ivanti says it is not a data store.
Ivanti describes two complementary paths:
- Native AI runs inside Ivanti products.
- MCP access extends Neurons context to supported external AI systems.
What data the AI can see
Ivanti's press release says the Neurons context available over MCP covers device, experience, patch, exposure and software intelligence. The configuration guide splits that into domain-specific servers:
| Domain | What it covers |
|---|---|
| Device | Inventory and health, OS and manufacturer breakdowns, warranty status, DEX scores, OData filtering |
| Data services | Natural-language questions converted into structured device queries |
| Patch | Compliance, deployment history, failures, trends, exploit severity, risk scores |
| Exposures | Exposure summaries, CVE details, severity |
| Software inventory | Installed software, versions, manufacturers, end-of-life status |
| Software distribution | Device and package deployment status, application catalogs |
| Reporting | Report configurations, jobs, execution history, distribution lists |
Ivanti's product page suggests uses such as patch investigation (finding failed patches and overdue scans), endpoint-health triage and fleet-health briefings. These are vendor examples. Ivanti has not published measured results.
Which AI clients are supported
Ivanti's configuration guide names three supported AI agent clients: GitHub Copilot CLI, GitHub Copilot and Claude. The client must also support OAuth-based MCP server connections. The broader product page adds Neurons agents. It also warns that support for other clients and custom agents may vary by release and should be confirmed in current documentation.
So "MCP-compatible" does not mean "supported." Test your chosen client before promising anything to your CISO.
Setting it up
Ivanti's documented procedure requires an active Neurons account with administrative access. It also needs a client ID from an AI Agent app registration. The steps are:
- Sign in to the Neurons tenant as an administrator.
- Go to Admin > Settings > AI Settings and toggle on Enable Ivanti MCP servers.
- Go to Admin > App Registrations > New Registration, choose AI Agent, and register.
- On the completion screen, copy the MCP-specific Neurons Auth URL and the client ID.
- Pick the MCP server domain that matches the task.
- In the AI client, enter the tenant URL and client ID, then finish authentication. Ivanti recommends OAuth.
Two practical details stand out:
- The URL pattern. The Auth URL must end in /mcp/<domain>/v1. Ivanti's example is a devices path on a placeholder tenant, not something to paste literally.
- Save the values immediately. Ivanti warns the client ID and Auth URL cannot be viewed again once the window closes. If you lose them, you must create a new app registration.
Governance claims, and how to read them
Ivanti says access is governed by OAuth, role-based access control, permissions and policy enforcement. It says Neurons keeps authority over the data. It also says it does not use customer data to train models.
These are vendor descriptions, not independent audits. Ivanti also notes that how an AI client handles returned data depends on the customer's agreement with that AI vendor. Data returned to Copilot or Claude leaves the Neurons boundary. Your review should cover that whole path, not just the Ivanti connector.
Choosing one domain per connection is a natural scoping lever. It lets you expose, say, patch data without handing over reporting or software distribution details. That is my reading of the documented design, not an Ivanti recommendation.
What it is not
The release page keeps three things separate:
- MCP Server access is GA and read-only.
- AI Insights & Recommendations is a tech preview. Ivanti says its initial experience surfaces prioritized issues and next steps but does not automatically execute them.
- Conversational AI Reporting is a Tech Preview (Tier 2).
The press release says administrators retain control over whether and how recommended actions proceed. Despite the "Autonomous Endpoint Management" branding, the GA MCP piece is about context, not action. The faster issue resolution in the headline is Ivanti's goal, not a demonstrated outcome.
Why Windows admins should care
Ivanti's Q4 release also lists day-zero readiness for Windows 11 26H2 in its Secure Access Client. That is a separate item from MCP. The MCP angle is that Copilot-based tooling can now query endpoint, patch and exposure data in an Ivanti estate through a standard connector. That could cut the effort of hand-building queries and custom integrations.
The sensible path is a pilot:
- Start with one low-risk domain, such as software inventory.
- Use a dedicated app registration.
- Confirm that existing RBAC limits what the AI client can see.
- Review your AI vendor's data-handling terms before widening access.
Treat the productivity claims as unproven until you have run it against your own fleet.
References
- Ivanti adds MCP connectivity to Autonomous Endpoint Management - UC Today UC Today · 2026-11-19T10:01:01.422000+00:00
- MCP Server for AI Governance – Ivanti Neurons | Ivanti ivanti.com
- GitHub - tyrunasj/ivanti-mcp-standalone · GitHub github.com