What Is AI Agent Security Gateway?
AI Agent Security Gateway is a security product that provides full lifecycle management and fine-grained security protection for MCP Servers. It helps customers inventory MCP Servers and API assets, discover vulnerabilities and risks in MCP Servers, identify and block attacks against MCP Servers, and reduce the risk of leakage of critical data assets.
How to Select a Model?
AI Agent Security Gateway currently supports only dedicated instance plans and does not offer shared instances. You can select a plan based on your business requirements, with a primary focus on QPS and bandwidth. To better ensure the stable and secure running of your business, we recommend that daily business metrics (QPS and bandwidth) remain below 50% of the plan's upper limit for an optimal experience.
What Are the Requirements for Activating AI Agent Security Gateway?
To deploy AI Agent Security Gateway, you need to select a VPC and subnet. If the existing network does not meet your requirements, you need to create a VPC. What Large Model Security Issues Can AI Agent Security Gateway Solve?
Traffic Risk Protection
Gateway traffic control: uses the "Gateway Traffic Control" feature of AI Agent Security Gateway to perform dynamic scheduling and monitoring based on parameters, versions, protocols, and loads.
Abnormal traffic detection: identifies and blocks abnormal access patterns, providing real-time alerts and blocking for abnormal requests.
Multi-dimensional traffic control: achieves fine-grained control over large model API calls through MCP Server routing configuration and full-dimensional traffic control capabilities.
Intrusion Risk Protection
Malicious prompt injection defense: identifies and blocks attack behaviors such as prompt injection, preventing attackers from bypassing security restrictions through carefully crafted prompts.
MCP Tool poisoning attack defense: detects and blocks malicious attacks targeting model-invoked tools, ensuring toolchain security.
Full-link security management: supports precise identification and handling of various security risks in large model scenarios such as input prompts, generated content, and runtime environments.
Business Risk Protection
MCP protocol encapsulation: provides standardized protocol conversion capabilities, supporting one-click encapsulation of traditional APIs into standardized MCP services for standardized API management.
Multi-version and multi-model dynamic scheduling: achieves dynamic scheduling of multiple versions, models, and tools through the gateway traffic control feature, reducing business risks caused by version confusion.
Content security review: identifies and blocks high-risk behaviors through the content review feature in the multi-dimensional protection system, preventing business data from being maliciously obtained.
Data Risk Protection
Sensitive information leakage prevention: includes built-in sensitive data identification rules to automatically mask high-risk fields such as ID card numbers and phone numbers.
Structured data security detection: performs security checks on model outputs to ensure compliance with security standards and prevent data leakage.
Security capability open integration: rapidly integrates third-party service components such as data security through an open security capability platform to build a complete data security protection system.
What Existing Product Solutions Can AI Agent Security Gateway Integrate with for Collaboration?
The AI Agent Security Gateway is now integrated with the content security operations platform to support moderation of content such as text and images, forming a multi-dimensional protection capability through capability convergence.
Does AI Agent Security Gateway Support Authentication-Free/Pure Proxy Mode?
Supported. When creating an application, select the No Authentication authentication type, and the associated model APIs and MCP resources can be accessed without credentials, enabling pure proxy forwarding. This is suitable for scenarios such as internal test environments and businesses with their own authentication systems. Note: In No Authentication mode, if other applications with configured authentication are associated with the same resources, their authentication policies will become invalid.
What Statistical Methods Are Supported for Token Control?
The cumulative consumption limit controlled by tokens supports three statistical methods:
Time window: a custom sliding time window (in minutes) within which cumulative consumption is counted, and after the window slides past, the count is automatically reset.
Natural day: Statistics are collected by natural day (00:00 - 23:59:59), and the counter is automatically reset at 00:00:00 (Beijing time) every day.
Natural month: Statistics are collected by natural month (from the first day to the last day of each month), and the counter is automatically reset at 00:00:00 (Beijing time) on the first day of each month.
What Happens to Existing Consumption Data After the Statistical Method for Token Control Is Switched?
After you switch the statistical method and save the changes, the system will reset the cumulative counter and start counting from a new statistical period. Existing consumption data will not carry over to the new statistical period.