Agent discovery & inventory

Every agent you run,found and owned.

One living inventory of every AI agent, copilot, MCP server and no-code flow, each with an owner and a scope.

The problem

You cannot govern
what you cannot see.

Agents arrive with the platforms you already use. Most aren’t malicious, just unreviewed, with access nobody tracked.

  • Built without askingNo-code flows wire agents into core systems before IT sees them.
  • Access nobody trackedStanding credentials and broad scopes, granted at rollout and never revisited.
  • No owner on recordNobody can say who approved an agent or what it should do.
A real pathA flow nobody registered starts writing to core.
  1. 1n8nA team builds a card-dispute automation in n8n and connects it to the core platform.
  2. 2core APIThe flow starts writing dispute records with a shared service key.
  3. 3ZenstraWorkflow Security maps the new flow: an agent with write access and no owner on record.
  4. HoldWrites to core wait until an owner and a scope are on record.n8n:card-dispute · write:core
Industry references
  • ASI10Rogue AgentsOWASP Top 10 for Agentic Apps
  • GOVERN 1.6AI system inventoryNIST AI RMF
  • ISO/IEC 42001AI management systemInternational standard
How Zenstra handles it

From discovery
to governed, in one place.

Discovery is only the first step. Zenstra connects what it finds to an owner, a policy and a record.

Discover

One inventory across every platform

Zenstra maps agents, tools, MCP servers and connections in workflows, in your cloud and on staff devices, and shows how much of each platform is governed.

  • Coverage by platform: governed versus discovered
  • Every hop in n8n, Zapier, LangGraph and CrewAI
  • AI on Windows and macOS, found by the AI Discovery app
Decide

An owner and a scope for each agent

Each agent gets an owner, a declared task and the grants it needs. Agents without an owner are flagged until someone takes them.

  • Highest-risk agents ranked, with the main factor
  • Unregistered agents flagged when they write to core systems
  • Agents without an owner shown on the overview
Prove

An inventory your examiner can read

The inventory is kept current automatically and exported as evidence, with the owner and the platform for every agent.

  • Living inventory, tracked automatically
  • Export as PDF and CSV
  • Every change to an agent on the decision record
Policy as code

The policy
behind it.

An agent counts as governed once it has an owner, a declared task and the grants it needs. Until then, policy decides what it may do.

  • Decided inline
  • Fails closed
  • Versioned with every decision
holdagent.*when owner is nonedenycore.writefor agents not in inventoryholdmcp.server.newuntil reviewedallowagent.runwhen owner and scope are set
What it finds

Agents hide in
six places.

  • Core-banking agents

    Agent runtimes inside the core platforms you already operate.

  • Employee copilots

    Microsoft Copilot, ChatGPT and Claude on staff devices.

  • No-code automation

    n8n, Zapier and Copilot Studio flows wired into production systems.

  • Agent frameworks

    LangGraph and CrewAI agents built by your own teams.

  • MCP servers

    Tool servers your agents call, approved or not.

  • Coding agents

    Cursor and Claude Code with access to source and secrets.

FAQ

Common questions.

How does Zenstra find agents nobody registered?

Three ways, by where they run: the gateway sees calls in your cloud, Workflow Security maps automation flows, and the Zenstra AI Discovery app finds AI on Windows and macOS devices.

Does discovery need agents to be rewritten?

No. Discovery and enforcement work with the agents you already have, through the SDK, the gateway or the AI Discovery app.

What happens to an agent once it is found?

It appears in the inventory, is checked against your policy straight away, and stays flagged until it has an owner and a declared scope.