MCP Gateway

The MCP Gateway is the PAS endpoint that AI Agents use to discover and invoke tools, prompts, and resources exposed by connected systems. It implements the Model Context Protocol (MCP) over HTTP/HTTPS.

The MCP Gateway serves two primary purposes:

  • Tool Discovery: Agents query the gateway to discover available tools, prompts, and resources across connected MCP services.

  • Tool Invocation: Agents call tools through the gateway, which routes requests to the appropriate backend service and handles authentication on behalf of the agent or user.

NOTE:

AI Agents require additional licensing. Contact your Imprivata Partner for more information.

Configure MCP Services for an Application

For an AI Agent to invoke tools exposed by a specific application, that application must have at least one host service configured with the MCP or MCPS port type. These port types route AI Agent traffic through the MCP proxy layer, which enables auditing and credential management.

User-Delegated OAuth for MCPs

Many external MCPs require authentication using OAuth 2.0 with user delegation. This flow allows users to authorize the system to access external services on their behalf, enabling MCPs to perform actions as that specific user.

The user-delegated OAuth flow follows the standard OAuth 2.0 Authorization Code Grant with Proof Key for Code Exchange (PKCE). This approach ensures security, user control, scope limitation, and revocation.

Personal Access Token Support for MCPs

Some external MCPs do not support OAuth delegation and instead require static authentication using personal access tokens (PATs). Your PAS server supports PAT authentication with flexible header configuration.

Personal access tokens provide a simpler authentication mechanism for MCPs that generate static tokens that do not expire (or expire rarely), authenticate using HTTP headers rather than bearer tokens, or require custom authentication headers beyond standard OAuth.

Manage MCP Consents

Users must explicitly authorize the VPAM server to access external MCPs on their behalf. The consent management interface allows users to view, authorize, and revoke these connections.

Audit and Compliance

All MCP activity is automatically audited for compliance and troubleshooting. The VPAM server tracks:

  • Authorization Events: When users authorize or revoke MCP consents.

  • Tool Invocations: Which tools are called, by whom, and when.

  • API Calls: Details of requests sent to external MCPs.

  • User Identity: Who initiated each action.

MCP Audit Logs are located in System AdministrationAdmin Log. Each MCP audit record includes:

  • Timestamp: When the activity occurred.

  • User: Who initiated the activity.

  • MCP Name: Which external service was accessed.

  • Tool/Method: Which specific tool or API method was called.

  • Parameters: (Limited visibility) Sanitized parameters used.

  • Status: Success, failure, or timeout.

  • Duration: How long the operation took.

Namespaced MCP Connections

In multi-site deployments, namespaces provide logical isolation of MCP connections. Namespaces ensure that users only see and can authorize MCPs relevant to their assigned site. A namespace is a logical grouping of related MCP services, typically aligned with organizational sites or departments.

When a user authorizes an MCP connection or views connected MCPs, the system resolves the user’s current namespace based on their assigned site or department, or the current context (web application, tool, etc.), resulting in the following:

  • Only MCPs configured for that namespace are displayed.

  • Authorization and revocation actions are scoped to that namespace.

  • Audit records include the namespace context.

MCP Gateway Reference

All requests to the MCP Gateway require a valid OAuth 2.0 bearer token issued for an AI Agent identity. See AI Agents for how to obtain a token.