AI agents are set to revolutionize how we work, create and compete, offering unprecedented efficiency and insight as they automate everyday tasks. But how do we ensure these intelligent allies are secure and can be trusted as they become integral to the ways we work and go about our lives?
At Palo Alto Networks, we believe that security must be a foundational enabler of AI adoption, not a barrier. That’s why we’re excited to introduce Prisma AIRS MCP Server, now available in public preview through sample code shared on PyPI and GitHub, so our customers can explore ways to secure the AI they build and deploy, including agents.
Scan Workflow Data for Malicious Content
This newly released public preview component of the Prisma AIRS platform helps protect the agents you build through an easy integration with Model Context Protocol (MCP), which is a standardized way to connect AI models to disparate data sources and tools.
This newly released public preview component of the Prisma AIRS platform can be integrated with your agent to detect malicious content in data. By connecting your agent, such as a cloud desktop, via the Model Context Protocol (MCP), you can immediately leverage Prisma AIRS to scan any text data.
This innovative solution is specifically designed to empower organizations to confidently build and deploy AI agents by embedding robust security directly into their agent design from the very beginning. The Prisma AIRS MCP Server can serve as an agentic interface for the AI Runtime API.
Open Standards (Like MCP) Accelerate AI Innovation but Don’t Solve Security
To accelerate the development of powerful AI agents, the tech industry is embracing open standards that allow different AI models, tools and data sources to communicate more easily with each other. One such important standard is the Model Context Protocol (MCP) – an open standard that creates a common language for AI models to talk to external tools, databases and services.
For developers, this is a game-changer. Instead of building custom integrations for each new AI model or data source, they can build to one standard. This means less duplicated effort, easier switching between AI providers, and faster development of more capable, context-aware AI applications. Learn more about MCP in this blog post.
This standardization is fantastic for innovation, because it reduces complexity and fosters a more open AI ecosystem. However, MCP doesn’t inherently address the security of the AI agents built using the protocol. Agents still require robust protection against the unique threats they face in the AI landscape.
See How to Embed Security into the Agentic AI Framework
This landscape is where the Prisma AIRS MCP Server comes into play. It acts as an intelligent security building block for AI agents utilizing the Model Context Protocol and has been designed to seamlessly integrate the advanced threat protection capabilities of Prisma AIRS directly into the communication flow of these agents.
This means that data passing through Prisma AIRS MCP Server are inspected in real-time, offering proactive defense against a wide array of AI-specific attacks. Features include:
- Connect with agents supporting MCP on any platform using any model, including AI agents in no-code, low-code and agent framework environments.
- Protection against AI agent-specific threats and 29+ prompt injection attack categories across eight different languages.
- Protection against web and DNS attacks – 40% better protection from web-based attacks and protection from 25+ DNS attack types.
- Custom topic guardrails and detection of harmful or toxic content.
- Data security with 1,000+ predefined data patterns and more.
- Automatic tool and feature updates.
Get an Early Start — Test the New Prisma AIRS MCP Server
The introduction of the Prisma AIRS MCP Server is a testament to our ongoing commitment to securing the future of AI. This server is a key addition to our comprehensive Prisma AIRS platform, ensuring security at every stage of the AI lifecycle.
This is your chance to be at the forefront of testing ways to secure AI agent adoption. The Prisma AIRS MCP Server, which can be run locally, is now available for public preview with published code on PyPI for customer-deployed testing. To fully leverage the open-source script, customers need access to the AI Runtime API. If you already have the AI Runtime API, you can use your existing API key.
Not a customer yet? Check out the interactive, self-guided tour to learn more or speak to your Palo Alto Networks account team to get a license today.