What should an MCP security review include?

It should include server inventory, tool permissions, command boundaries, authorization, data paths, secrets handling, logging, recovery, and prompt/tool injection scenarios.

Is this only for production systems?

No. It is better to review MCP and tool-connected agents before production, but existing systems can also be tested and remediated.

Aeon AI Risk Management

MCP security review checklist for tool-connected AI.

Aeon reviews MCP servers and tool-connected agents across inventory, permissions, command boundaries, authorization, secrets, logs, data paths, and recovery before deployment.

Questions this page answers

What should an MCP security review include?: It should include server inventory, tool permissions, command boundaries, authorization, data paths, secrets handling, logging, recovery, and prompt/tool injection scenarios.
Is this only for production systems?: No. It is better to review MCP and tool-connected agents before production, but existing systems can also be tested and remediated.

Inventory the tools before reviewing risk

List MCP servers, tool handlers, commands, APIs, data stores, credentials, execution contexts, and the agents or users allowed to invoke them.

Test the command and authorization boundary

Review role enforcement, tenant boundaries, input handling, command construction, secrets exposure, prompt/tool injection, logging, and rollback behavior.

Close with remediation evidence

A checklist is useful only if findings become fixes. CyberGuard verifies issues, maps them to developers, and retests closure where scope includes validation.