Integration: Cisco AI Defense Plugin for Google ADK
What it does
Cisco AI Defense is a Google ADK plugin that integrates runtime LLM and MCP tool inspection directly into the ADK Runner lifecycle using a BasePlugin and per-agent callbacks.
It inspects user prompts before model calls, model outputs after generation, and MCP tool call requests/responses — then allows or blocks based on configurable security policies. No monkey-patching required.
Key capabilities
- Runtime protection for Gemini calls: Inspect user prompts before model calls and model outputs after generation, then allow or block based on policy (Monitor or Enforce).
- Tool and MCP call inspection: Inspect tool call requests before execution and tool responses after execution, and block unsafe tool behavior in Enforce mode with clear metadata.
- Auditable decision trace and alerts: Capture decision context (action, severity, classifications, request_id/event_id) and optionally trigger an
on_violation callback for monitoring and incident response.
Integration type
- Plugin (
BasePlugin subclass) — global, applies to all agents in the Runner
- Agent callbacks — per-agent, wired via
before_model_callback, after_model_callback, before_tool_callback, after_tool_callback
Installation
pip install cisco-aidefense-google-adk
Links
Proposed documentation
I'd like to add an integration page at docs/integrations/cisco-ai-defense.md following the integration contribution template, including:
- Frontmatter with catalog metadata
- Use cases, prerequisites, installation
- Working code examples (plugin and callback approaches)
- Modes table (monitor / enforce / off)
- Callbacks covered table
- Resource links
A logo asset will be included at docs/integrations/assets/cisco-ai-defense.png.
I will follow up with a PR once this issue is acknowledged.
Integration: Cisco AI Defense Plugin for Google ADK
What it does
Cisco AI Defense is a Google ADK plugin that integrates runtime LLM and MCP tool inspection directly into the ADK Runner lifecycle using a
BasePluginand per-agent callbacks.It inspects user prompts before model calls, model outputs after generation, and MCP tool call requests/responses — then allows or blocks based on configurable security policies. No monkey-patching required.
Key capabilities
on_violationcallback for monitoring and incident response.Integration type
BasePluginsubclass) — global, applies to all agents in the Runnerbefore_model_callback,after_model_callback,before_tool_callback,after_tool_callbackInstallation
Links
Proposed documentation
I'd like to add an integration page at
docs/integrations/cisco-ai-defense.mdfollowing the integration contribution template, including:A logo asset will be included at
docs/integrations/assets/cisco-ai-defense.png.I will follow up with a PR once this issue is acknowledged.