Security Hub MCP makes agents auditable

AWS has introduced the Security Hub MCP App in preview. It brings exposure findings from AWS Security Hub into Claude Desktop through a local Model Context Protocol server. For enterprises, this is more than a convenience feature: security analysts can investigate findings, attack paths and remediation guidance in natural language without constantly switching context.

What changes

According to AWS, the MCP App can show top exposures, details for individual findings, correlated findings, affected resource configurations, attack paths and remediation recommendations. Each tool call provides two layers: a text summary for the agent to reason over and an interactive visualization that the human analyst can verify in the same conversation.

The operating boundary matters. The documentation describes the app as a local MCP server that uses existing AWS credentials to call Security Hub APIs directly. The available tools are read-only and, according to AWS, do not modify the security posture. The preview is available in all commercial AWS Regions where Security Hub is available.

Why it matters

MCP is moving from the developer workbench into security operations. An agent no longer only answers text; it can access real cloud findings, resource relationships and prioritization signals. That is where governance starts: Which identity does the MCP server use? Which findings may the analyst see? Which visualization counts as auditable evidence? Which recommendation remains advice, and which one becomes a ticket?

OWASP describes Excessive Agency as a risk when LLM systems receive too much functionality, permission or autonomous scope. AWS addresses part of this through read-only tools. For production use, that is not enough on its own: logs, role models, data classification and human approval remain mandatory.

DACH perspective

For DACH organizations, this pattern is relevant because security teams often struggle with alert volume, tool switching and scarce specialist capacity. At the same time, auditability, privacy and incident processes must not disappear into a chat window.

A pragmatic starting point is an MCP operating profile: approved AWS accounts, IAM roles, data classes, logging, export rules, ticket handoff and mandatory review. That turns a security agent from an autonomous black-box helper into an auditable analyst layer.