AI Pulse · Under the Radar

Blitz Strike is a new open-source toolbelt aimed at testing MCP environments for security weaknesses. It lists 57 escalation chains and a catalog of 130 tools, which suggests its focus is not just finding one-off issues, but testing how access, permissions, tools, and agents can be combined in ways teams did not intend.

It is also gaining attention quickly, with 637 GitHub stars at a reported rate of 87 per day. That does not prove the tool is mature or safe to use in every environment. It does tell you that more security practitioners are looking closely at a problem many businesses have barely started to define, how to audit AI agents and the MCP connections that give them access to company systems. You can review the project details in the original repository.

For an operator, the practical question is simple. If an AI assistant can read files, query data, send messages, or call internal tools, who has tested the paths between those actions? Traditional application security checks may not cover an agent that can chain several approved actions into an unapproved outcome.

This is the kind of thing we build into an AI command centre, visibility over what AI systems can access, what they are doing, and where human approval should sit. Security auditing of agent and MCP surfaces is starting to look like its own discipline, not a small extension of existing IT controls.