The State of MCP Server Security in 2025
In this CSA CloudBytes webinar, Jonathan Sander, Field CTO at Astrix Security, and Tal Skverer, Head of Research, reveal what more than 5,000 open-source MCP servers tell us about the true security posture of AI agent infrastructure. As MCP becomes the default way to connect LLMs and agents to APIs, it is quietly creating a new non-human identity attack surface filled with static secrets, agent swarms, and blind spots most security teams are not prepared for.
Drawing on large scale ecosystem analysis and a global survey of practitioners, they break down how MCP is actually used in the wild, why 88% of servers require credentials, and how those credentials are typically stored in plain text on developer machines. They also introduce the open source MCP Secret Wrapper, a practical way to pull secrets out of local configs and into vaults, so organizations can start treating MCP as an AI agent control plane instead of a new shadow risk.
Watch the webinar on-demand to learn:
The real state of the MCP ecosystem in 2025
See what thousands of MCP servers and hundreds of survey responses reveal about adoption trends, where servers actually run, and how quickly AI agents are driving an identity explosion.
How MCP servers handle identities and secrets today
Understand how often servers rely on static keys embedded in configs, why so few use OAuth, and how this pattern expands your non-human identity attack surface for infostealers and lateral movement.
How to turn MCP into a control plane
Learn how the open source MCP Secret Wrapper lets you source credentials from vaults like AWS Secrets Manager, reuse existing MFA and SSO controls, and create a safer foundation for AI agent governance without blocking developers.