Salt Security Embraces MCP to Improve Cybersecurity in the Age of AI
Salt Security this week at the 2025 RSA Conference made available an early preview of an ability to secure Model Context Protocol (MCP) servers that are emerging as a de facto standard for integrating artificial intelligence (AI) models and agents.
At the same time, Salt Security will make the security intelligence it collects available to other cybersecurity platforms using MCP.
Salt Security CEO Roey Eliyahu said MCP, at its core, is essentially a type of application programming interface (API) that the company’s core platform can be extended to support. That’s critical because if an MCP server is compromised, it does not just lead to, for example, data being exfiltrated. Instead, it can potentially result in entire workflows and processes being hijacked because AI agents have been created to perform an action, he added.
Capabilities that Salt Security will enable via an MCP server include contextual search across an entire inventory of APIs, tools that identify gaps and misconfigurations, and an API explainer that, in addition to making it simpler for understanding how an API functions in natural language also enables the Salt Security platform to offer API remediation guidance.
AI agents have yet to be pervasively deployed at scale, but it may now only be a matter of time before they are compromised via an MCP server. Hopefully, cybersecurity teams will proactively address these cybersecurity concerns, but if history is any guide,e they may once again be reacting to attacks they are at the moment unprepared to thwart.
On the plus side, however, those same MCP servers will make it easier to share security intelligence across multiple tools and platforms that have been infused with AI. Originally developed by Anthropic, MCP provides two-way connections between data sources and AI tools. Cybersecurity and IT teams can both expose data through MCP servers or build AI applications, also known as MCP clients, that connect to these servers. That approach makes it possible to query internal systems without blindly scraping data or exposing backend systems. In effect, an MCP server acts as an intelligent gateway, translating natural language prompts into authorized, structured queries while enforcing security, governance and usage policies.
It’s not clear to what degree the rise of agentic AI might serve to bring more attention being applied to API security, but it’s an area that has long not received enough attention, even though every modern application exposes one. Historically, API security has been long considered a subset of application security, which many cybersecurity teams have assumed that application development teams are addressing. Application developers, conversely, assume that cybersecurity teams are addressing those concerns, resulting in neither team being especially focused on either applications or the APIs that support them.
AI agents will compound that issue because they continuously invoke MCP servers, resulting in more API interactions that will need to be secured.
It may be a while before MCP servers are pervasively deployed, but it’s now more a question of not so much whether they will be deployed as much as it is how many will need to be secured.