API and AI agent security
An API or an AI agent can't be scanned the way a repository can. You have to talk to it. And an agent that can call tools is a different risk from one that can only answer questions, so it matters what it can do as well as what it says.
These two resource types describe the surfaces you expose and whether anyone actually tested them.
This page describes a preview. The screens exist in the console, but connecting these resources and testing them isn't generally available yet. What you see can change before release.
What each type holds
| Type | Kinds | Facts it tracks |
|---|---|---|
| API Security | API, web app | Domain, TLS, authentication, spec, endpoints, method |
| AI Agent Security | Agent, MCP server | Model, tools, guardrail, transport, authentication |
Tested means tested
Other types say scanned. These two say tested, and the distinction is the point of the page. A surface can be found without ever having been exercised, and a list of APIs nobody has tested is a list of unknowns. The console counts tested and untested separately so you can see the gap.
The Connector tab
These types add a Connector tab, right after Lineage. A connector is the saved recipe for reaching the application: where it lives, how to authenticate, what a request looks like and where the answer sits in the response.
Tests depend on it. Without a working connector, the application can be listed but not tested.
How it fits with the rest of the console
Agents link back to the code they're built from, so a weakness found by testing an agent can be traced to a repository once lineage is mapped. See Lineage.
Findings about AI code that you can get today come from source scans. The AI code & MCP config category and the Skills & agent instructions category both run on repositories. See Code Security.
Try this next
You don't have to wait for the preview to learn what agents you have. Open Components, go to the AI group, and see which models, agents, tools and MCP servers your code and images already contain.