MCP Servers

    MCP Servers in ManageEngine SecureAI provides a consolidated inventory of the MCP (Model Context Protocol) servers that an organization's AI agents are connected to, spanning across all onboarded AI platforms. It reports which of those connections the organization's policies already cover and which remain reachable without them.

    Administrators use MCP servers to establish how many of their agents' MCP connections are governed and to prioritize remediation for the servers that are not. The inventory is read-only. Remediation is configured in Access Protection and completed in the AI platform that owns the agent.

    How MCP servers are discovered and protected

    The MCP Servers feature holds no records of its own. Every server listed comes from the application syncs configured under Application Settings. A sync collects the endpoints that the instance's agents are configured to call and lists each one as an MCP server. A server is therefore listed only when at least one discovered agent references it. The inventory reflects each instance's last successful sync, which runs every six hours by default. Selecting Fetch against an instance under Application Settings pulls a fresh set immediately.

    An MCP gateway publishes a single URL that agents call in place of a server's own address and forwards those calls to the servers behind it. A server is counted as protected once it sits behind a gateway and its agents call that URL. What the gateway then allows is set by the control policies attached to it. The endpoint URL is held by the AI platform rather than by SecureAI, so protection is completed in two places. Registering the server and binding it to a gateway in SecureAI builds the path. Replacing the agent's endpoint URL with the gateway URL in the AI platform puts that path into use. The status changes after the next sync reads the updated endpoint.

    Prerequisites

    • The technician reviewing the inventory must have viewing access to the AI inventory. The default Governance Manager role carries this permission.
    • The technician who protects a server must have permission to configure gateways and their policies, which the default Governance Manager role does not carry.
    • Must have administrative rights in the platform that owns the agent since the endpoint URL is replaced on the agent itself.
    • Must add each AI platform instance under Application Settings and associate the relevant service with it.
    • Must sync each instance and confirm that the sync has completed.
    • Must configure an identity source. A gateway uses this to validate the tokens it receives.
    Note:
    • A discovered MCP server cannot be added, renamed, amended, or removed.
    • Servers cannot be protected in bulk. The Configure link for an unprotected server acts on that server alone.
    • The MCP Server column cannot be removed from the display.
    • Changes reach the inventory only at the next sync. A server connected moments ago, a server no longer in use, and an endpoint URL that was already switched to a gateway are all reported as they stood at the last sync.
    • Servers that no discovered agent references and servers on AI platforms that have not been onboarded are absent from the inventory.
    • A protection status of Protected by gateway does not confirm that the gateway can currently reach the server.
    • The protection status does not indicate what the gateway allows or blocks. A gateway cannot carry control policies on its own.

    Configuration instructions

    1. Log in to SecureAI and navigate to Governance > AI Inventory > MCP Servers.
    2. Review what the inventory reports for each server: the endpoint URL its agents call, the application and instance that discovered it, the agents configured against it, and its protection status.
    3. Select the filter icon above the inventory and choose All, Unprotected, or Protected to restrict the inventory to the servers with that protection status.
    4. Only servers reported as Unprotected carry a Configure link. Select it to display the Protect MCP Server pop-up, which names the server and the two changes required to protect it.
    5. Select Configure in Protect MCP Server pop-up to continue to Access Protection, where the server is registered and bound to a gateway. Once that is done, the server is reported as Protected by gateway from the next sync onwards.

    Tips

    • Place servers that serve the same team or that carry the same level of risk, behind one gateway. The control policies attached to that gateway then cover every server behind it, and the owner of the AI platform has one URL to apply. See MCP Gateways.
    • Protection takes effect only when the agent is redirected at the gateway URL, which is usually done by another team in the AI platform. Schedule that change with them, then check Gateway Activity to confirm the agent's calls are reaching the gateway.
    • Reviewing the inventory and configuring gateways usually fall to different teams. Instead of widening one technician's permissions to cover both, assign the reviewer to the Governance Manager role and route the protection work to the technician that holds Access Protection.
    • If the status has not changed after the agent was updated, check that the instance has synced before examining the gateway. Connection Status under MCP Server Registry reports whether the gateway can reach the server.