Shadow AI discovery

See the AI your network never sees

Ferentin inventories the AI agents, MCP servers and plugins on every managed Mac. Each finding ends in one next step: govern it, approve it or quarantine it.

The problem

A stdio MCP server never touches the network

An engineer adds a local MCP server to Cursor and gives it a token. It runs as a subprocess speaking over pipes and never opens a socket, so no proxy, secure web gateway, CASB or DNS filter ever sees it. Coding agents, local models and plugins arrive the same way: installed by a developer and missing from every network-only inventory.

How it works

From unknown to governed in four steps

  1. 1

    Put the agent on every Mac

    Deploy the Ferentin endpoint agent through Jamf Pro, Intune or any MDM that can push a signed package and profiles. The report states its own denominator, such as found on 42 of 48 enrolled devices, so you know how much of the fleet you are looking at. Devices that stopped reporting are flagged rather than read as clean.

  2. 2

    Inventory every AI agent

    Claude Code, Cursor, local model runtimes and the apps that host them are found by process, bundle, config file and network flow. Each row shows device counts, versions and the code identity of the running binary, from Verified to Unsigned. A tool found more than one way is harder to miss.

  3. 3

    Find the MCP servers no proxy can

    The agent reads each AI client’s MCP configuration on disk and reports every server it declares, local or remote. You see each server’s name, the file that declares it and the names of the environment variables it expects, never their values or its arguments. Remote servers are matched against your registry and the Ferentin catalog.

  4. 4

    Turn each finding into a decision

    Every row ends in one next step. Govern opens an Endpoint Policy rule with the tool already selected. A quarantined MCP entry is rewritten to explain the block, never deleted, so the developer can ask for it back.

Works with your stack

Works with your stack

Discovery covers the coding agents, IDEs and local model runtimes your developers install. The agent deploys through the MDM you already run.

Browse all integrations

FAQ

Shadow AI Discovery questions

Not covered here? Ask us.

Shadow AI is AI software people use without the security team’s knowledge or approval: a coding agent running on a personal API key, a local MCP server added to a config file, a plugin installed into an IDE. It is the successor to shadow IT with higher stakes, because these tools read source code and customer data and act on the user’s behalf. Some of it shows up as network traffic. Much of it never does.

See it on your own stack

Walk through it with our team on your identity provider, your models and your MCP servers. Or start free today.