Visibility | AI Agent Security
Most AI agents on your endpoints were never formally deployed. A developer installs a coding assistant, it connects to an MCP server and an LLM provider, and a marketplace skill changes how it behaves, none of which appears in a traditional asset inventory. Visibility is where AI Agent Security records what it found, so you can work from a list of what is actually running rather than what you assume is running.
Scope of this guide
This guide introduces Visibility and what each page in it covers. For the columns, filters, and actions on an individual page, see the page's own guide.
Prerequisites
- You have administrator access to the tenant.
- The AI Agent Security agent is installed on at least one endpoint.
What Visibility covers
Visibility has five pages, each answering a different question about your environment.
| Page | What it lists | The question it answers |
|---|---|---|
| AI Agents | Every AI agent discovered, with the endpoints it runs on, the MCP servers on those endpoints, and the activity it launched | What agents are in use, and how far has each one spread? |
| MCP Servers | Every MCP server discovered, with how it connects and whether it requires authentication | What can those agents reach, and which servers accept connections without authentication? |
| LLM Endpoints | The LLM destinations your endpoints connect to, with the devices and processes that connected to each | Which LLM providers and gateways are your agents sending data to? |
| Agent Skills | The agent skills found on your endpoints, with their risk | What instruction files are changing how agents behave, and which ones need review? |
| Endpoints | Every machine running the AI Agent Security agent, with its status and the policies attached to it | Which machines are reporting, and what governs each one? |
How Visibility is useful
It shows you what no one declared. Agents, servers, LLM destinations, and skills appear here because they were discovered on your endpoints, not because someone registered them.
It connects the views. An agent's details lead to the endpoints it runs on and the MCP servers and LLM endpoints it talks to. An endpoint's details lead to the agents, MCP servers, and skills found on it. You can start anywhere and follow the chain.
It tells you where you are exposed. A server that accepts connections without authentication, a widespread agent with no policy in effect, or an endpoint that stopped reporting is visible from a list before it turns into an incident.
It scopes a change before you make it. Policies attach to endpoints, so blocking an agent or restricting a server affects every endpoint the policy reaches. Visibility shows you what that is first.
Verify the results
Confirm that the counts on each page match what you expect for your fleet. When a search or filter is active, the count reflects only the matching items, so check for active filters before treating a low number as real.
Next steps
- Review the AI Agents page and confirm that a policy governs each widespread agent.
- Review the MCP Servers page for servers that do not require authentication.
- Review the Endpoints page for machines that stopped reporting.
Updated about 2 hours ago
