Microsoft Defender Maps Local Agents, MCP Servers
Microsoft Defender for Endpoint gives security teams activity-based visibility into local AI agents and their configured MCP servers across supported endpoints. Its preview runtime protections can inspect agent activity through vendor hooks or network inspection.
Defender is treating local AI agents as a new endpoint-permission problem, and that is the right framing: the real risk is not which assistant is installed, but what files, identities, and tools it can reach.
- –Discovery covers major coding agents, agentic IDEs, desktop assistants, and local runtimes including Claude Code, Codex CLI, Cursor, Ollama Desktop, and LM Studio.
- –MCP visibility exposes the servers that give agents operational reach, turning opaque configuration files into security telemetry.
- –Activity-based discovery means installation alone may not create an inventory record, so dormant or rarely used agents can remain invisible.
- –Runtime protection is narrower than discovery, relying on supported hooks or network inspection and remaining in preview.
- –The feature is highly relevant for developer workstations, but Plan 2 licensing and endpoint coverage will determine how broadly enterprises can deploy it.
DISCOVERED
1h ago
2026-10-21
PUBLISHED
1h ago
2026-10-21
RELEVANCE
AUTHOR
DIY Smart Code