Securing Toolbox with Model Armor
Protect your agents and tools against prompt injection and sensitive data leakage by screening traffic with Google Cloud Model Armor.
Security
Harden your MCP Toolbox agents and tools against prompt injection, jailbreaks, and sensitive data leakage.
less than a minute
This section covers how to secure MCP Toolbox deployments against common AI security risks like prompt injection, jailbreaks, unauthorized database modifications, and sensitive data leakage across the traffic between your users, agents, and tools:
- Model Armor: Defend AI applications against prompt injection, jailbreaks, and sensitive data leakage by screening traffic with Google Cloud Model Armor.
- Read-Only Tools: Enforce deterministic read-only database access across custom tools and prebuilt servers using engine-level protocol locks, tool suppression, and MCP annotations.
Read-Only Tools
Enforce deterministic read-only database access across custom tools and prebuilt servers using engine-level protocol locks, tool suppression, and MCP annotations.
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.