AI agents should not use MCP server tools without review
Why This Matters
Copilot Studio agents integrated with external MCP servers can introduce supply chain risks if those endpoints are compromised or untrusted. An attacker who controls an MCP server could manipulate agent behavior, exfiltrate data, or execute unauthorized actions. Reviewing all MCP server integrations is essential to maintain control over your AI agents and protect your organization from third-party vulnerabilities.
What Aether365 Checks
This check verifies whether any Copilot Studio agents in your tenant have MCP server tool integrations that have not been reviewed for security compliance. It appears in the Aether365 dashboard under the "microsoft-365" category with severity Medium.