Ferentin for Endpoints

AI governance that lives on the Mac

A managed macOS agent you deploy through Jamf Pro or any MDM. It inventories AI tools, governs MCP servers and checks agent tool calls against a signed policy that keeps working offline.

What it covers

What it sees on every Mac

The agent reports what is installed, configured and running, with evidence for each finding. It does not read documents, capture keystrokes or inspect the contents of network traffic.

  • AI agents and local runtimes

    Claude Code, Cursor, Codex, GitHub Copilot, local model runtimes and the apps that host them, found by process, bundle, config and network evidence.

  • MCP servers, local and remote

    Every server declared in each AI client’s config, including stdio servers that never open a socket. Never their arguments, environment values or headers.

  • Skills, plugins and hooks

    What is installed into each agent, identified by a hash of its files. A changed version under a familiar name stands out.

  • Where AI traffic goes

    On the network filter track, each flow is attributed to the signed app that opened it, by signing ID and Team ID. No TLS interception.

  • Device posture

    System Integrity Protection, FileVault, boot security, content filter liveness and more, reported on a five-minute heartbeat.

  • Device identity

    A P-256 key in the System Keychain and a short-lived mTLS certificate. Each Mac is a security principal of its own, separate from the person using it.

How it deploys

Deploys like every other Mac agent

A signed, notarized package and a set of configuration profiles, pushed from your MDM. No user-facing installer, no license key and no local config file to edit.

  1. 1

    Publish from Ferentin into Jamf Pro

    Give Ferentin a Jamf Pro API client and it uploads the agent and removal packages, registers extension attributes, mints the enrollment token into the enrollment profile and scopes an install policy to your smart group. You upload the signed profiles yourself, because Jamf’s API cannot accept them. Jamf Now, Intune, Kandji, Mosyle and Workspace ONE each have a written procedure.

  2. 2

    Each Mac enrolls and reports

    The agent enrolls with a device key and a short-lived certificate, then reports posture on a five-minute heartbeat. Each Mac’s details show which security checks pass and which coding agents have Ferentin’s hooks in place. Start with a pilot group, then move through Internal, Canary and Production rings.

  3. 3

    People sign in once

    The person at the Mac signs in to the Ferentin app through your identity provider. From then on the app tells them what Ferentin did and why: a prompt blocked on a credential, a tool call a rule denied, an MCP server held for approval. It never shows the matched text, because that text never reaches the app.

What it enforces

One signed policy, enforced on the device

Policy arrives as a signed bundle and is verified on the Mac. A Mac that cannot reach the control plane keeps enforcing its last verified policy.

  • MCP server rules

    Allow, steer or block each server. Approved remote servers are re-pointed at your Ferentin gateway. A blocked entry is rewritten to explain itself, never deleted, so the developer can appeal.

  • Local MCP servers through one launcher

    Approve a local server once for the fleet. It starts through ferentin-mcp, which checks policy before the process runs and governs every tool call.

  • Tool call rules

    Block a single command such as curl or a whole tool. On Claude Code and Cursor a match is denied when the scope is set to Deny. Elsewhere it is counted, so you can measure first.

  • Data protection on the device

    Your Data Protection policies scan tool calls, prompts and tool output on the Mac. A call carrying a blocked secret is denied before it runs. Matched text never leaves the device.

  • Network destinations

    Allow or block AI destinations, pinned to an app by bundle ID, signing ID or Team ID. Available on the network filter track.

  • AI agent decisions

    Decide per cataloged AI agent and device group: Allow and report or Block. A blocked agent is refused on managed Macs.

Works with your stack

Works with your stack

Deploy through the MDM you already run. The agent governs the coding agents your developers already use.

Browse all integrations

FAQ

Endpoints questions

Not covered here? Ask us.

Ferentin for Endpoints is a managed macOS agent for enterprise AI governance. It inventories the AI agents, MCP servers and plugins on every Mac, attributes where their traffic goes and enforces a signed policy that keeps working when the network is down. It is deployed like any managed security agent: a signed, notarized package and configuration profiles pushed from your MDM.

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.