Frequently Asked Questions About Agentic Identity

How does Island tie an agent's activity back to a person?

Island logs each event with the authenticated user, the agent or application, and the session wherever the source exposes them, and it links the event to the responsible person whenever it can determine that link. The record keeps a person's own actions separate from the actions their agents take on their behalf.

How does Agentic Identity work alongside Okta, Microsoft Entra, and PAM tools?

Alongside them. Identity providers and privileged access management (PAM) tools govern the access path: which identity may reach which resource, with what credential, for how long. Island governs what the identity does at runtime. Island integrates with Okta and Microsoft Entra ID and uses their users and groups in policy, including MCP tool profiles. Organizations can apply stricter AI policy to privileged groups, with the identity provider or PAM system remaining the source of truth for who counts as privileged.

What is the Island MCP Gateway?

The Island MCP Gateway is a gateway hosted by Island that sits between agents and the remote MCP servers an organization approves, including internal servers that stay private behind Island Private Access. Agents send their MCP traffic through it, and the gateway inspects each tool call before forwarding it, using the tool name, its arguments, and the user and agent behind the request.

Which MCP servers can agents reach through the gateway?

Only the servers administrators approve. The MCP Gateway works from an approved catalog and denies routed connections to anything outside it by default. Administrators can also distribute the approved server configurations to supported MCP clients.

Can the gateway limit an agent to certain tools or to read-only access?

Yes. The gateway lists the tools each routed server exposes, and tool profiles assigned to users and groups decide which of those tools an agent may call. A profile can hold an agent to a read-only or non-destructive set, and inline policy can deny a single call based on the tool and its arguments. When the gateway denies a call, supported MCP clients receive a structured block response or tool feedback.

Does Island give AI agents just-in-time access?

Yes. The Island MCP Gateway issues a credential scoped to a single task that expires when the session ends. The agent authenticates to the gateway and the gateway authenticates onward, so the raw credential never reaches the agent. Blocking an agent at the gateway cuts off its access without rotating the underlying key.

Which credential does the gateway use to reach a downstream application?

Either a credential stored in the gateway's vault or the invoking user's own OAuth connection to the application. The gateway injects that credential itself, so the agent does not need to hold the underlying secret. Downstream permissions follow the credential in use: a user's OAuth connection applies that user's permissions, and a vault credential applies its own.

How do we cut off an agent's access quickly?

Administrators can restrict a user, an agent, a tool, a server, or a gateway credential from one place, and the change applies to every later request that passes through the MCP Gateway. Because the gateway holds the credential, blocking an agent there leaves the underlying key intact for everything else that uses it. Removing a user or changing their group in the identity provider also narrows their access through Island once the change propagates.

How does Island keep agents on the governed path?

Application and network policy steer agent traffic to the gateway, and providers that support allowlists can accept only the gateway's addresses. Credential brokering adds a third control: an agent that never holds a key cannot use it anywhere else. Restricting other outbound routes and retiring credentials held outside the gateway closes the remaining gaps.

Does the gateway govern agent-to-agent calls?

When one agent reaches another as an MCP server through the gateway, Island treats the call as MCP tool traffic, so the same inspection and tool policy apply to it.

What does the gateway record, and where can the events go?

Each gateway event records the user and agent where available, the server, the tool, the arguments, the result, and the policy verdict, subject to the organization's capture settings. Those events export to security information and event management (SIEM) platforms along with the rest of Island's audit data.