Registry
The Registry is your complete, real-time inventory of every AI agent in your organization showing what each agent connects to, who owns it, and how risky it is.
What is Agent Registry
The Registry is the central discovery hub in Agent Identity Security. When you connect your AI platforms through Integrations, Agent Identity Security automatically finds every agent running in your environment and records it here.
Think of it as your AI Bill of Materials (AI-BOM). Just as a software BOM documents every component in a codebase, the Registry documents every AI agent regardless of which platform runs it, which team deployed it, or when it was created. Every Agent Identity Security capability such as risk scoring, posture analysis, policy enforcement, compliance monitoring is built on Registry data.
The Registry surfaces not just the agents, but everything they rely on:
- The MCP servers they connect to.
- The AI models they run on.
- The skills they use.
- The people who own and access them.
- The credentials they carry.
What the Registry shows you
The Registry answers the questions that matter most when you are trying to understand and secure your AI footprint:
| Question | Where to look |
|---|---|
| How many agents do we have, and where are they running? | The Agents list. You can filter it by platform, status, and risk level. |
| Which agents are the riskiest, and who owns them? | The Agents list sorted by Risk Level. The Owner column shows who is responsible. |
| What external tools and services do our agents connect to? | The MCP Registry. Shows every MCP server connected to agents, with sanctioned or unsanctioned status. |
| Which AI models are in use, and are they approved? | The Models Registry. Shows every model in use with approval status. |
| Who is accessing agents, and are any agents left without an owner? | The Users view. Filter to show agents where Owner is empty. |
| Are all credentials still active, or are some abandoned? | The Credentials view. Shows last-used date and active status for every credential. |
What the Registry tracks
The Registry is organized into six views. Each covers a different part of your AI environment:
| Registry view | What it contains |
|---|---|
| Managed Agents | Every AI agent discovered across all connected platforms such as Microsoft Copilot Studio, Claude Code, n8n, GitHub Copilot, Salesforce Agentforce, and more. Each entry shows the agent's platform, risk level, owner, and status. |
| Unmanaged Agents | Agents detected through endpoint activity that are not registered or managed through any connected platform. Each entry shows the agent's host device, risk level, owner, and status. |
| MCP Servers | All MCP servers connected to your agents. Each server is marked as Sanctioned or Unsanctioned. Unsanctioned servers are a leading cause of elevated risk scores. |
| Models | Every AI model your agents are running on. Shows whether each model has been approved by your organization through the Models Registry. |
| Skills | The capabilities and tools attached to each agent. Helps you understand what each agent is able to do. |
| Users | The people who own and access agents. Use this view to find agents with no identified owner those need ownership assigned before they can be properly governed. |
| Credentials | API keys, tokens, and other credentials associated with agents. Shows which are active and which have not been used recently stale credentials are a security risk. |
How the Registry connects to the rest of Agent Identity Security
The Registry is the foundation that every other Agent Identity Security feature depends on. Without a complete, accurate agent inventory, risk scores, posture checks, and compliance reports are incomplete.
- The Agent Risk dashboard uses Registry data to score and rank agents by risk.
- The Agent Posture dashboard uses Registry data to surface configuration issues across the fleet.
- The Govern section uses Registry data to apply policies and track remediation.
- The Comply section uses Registry data to map agents and issues to regulatory frameworks.
From Registry to action
The Registry is read-only, so every security action starts with something you find here and then follows a path to Govern or Secure depending on whether you want to investigate a problem or prevent one.
| What you find in Registry | Where to go | What to do |
|---|---|---|
| Agent with a Critical or High risk score | Govern — Event Logs and Agent Posture | Review the events that raised the score, then open the specific posture finding to identify what is wrong. |
| Agent connecting to an unsanctioned MCP server | Secure — Policies | Create a Deny rule targeting that MCP server. Optionally add a Gateway endpoint to audit or scan the traffic before blocking. |
| Agent with no identified owner | Govern — Agent Posture or Workflows | Find the missing-ownership posture finding and assign an owner, or set up a Workflow to alert your team when new ownerless agents appear. |
| Credentials that have not been used recently | Govern — Agent Posture | Stale credentials generate posture findings. Open the finding to review last-used dates and decide whether to revoke. |
| AI model with no approved status | Govern — Compliance | Check which compliance controls the unapproved model affects, then approve or block the model through the Models Registry. |
| Agent inactive for 90 or more days | Govern — Workflows | Create a Workflow triggered by the inactivity event to flag the agent for review or initiate decommissioning automatically. |
| Agent connecting to high-risk external tools | Secure — MCP Gateway | Add a Gateway endpoint so every request is authenticated, scanned for PII or embedded secrets, and recorded in Event Logs. |
Keep the Registry accurate by reviewing it regularly especially after new integrations are connected or new teams start deploying agents.
