Skip to content

Every tool in your stack sees part of an agent. Here’s where each one stops.

We give each tool full credit for what it sees. Then we show where it stops, on the laptop, in SaaS, and in the cloud, and where Neo picks up.

Three nested frames: the network, the device, and the inside of the tool. Inside the tool, Claude Code reads a file, runs a shell command and calls an MCP server on the laptop; none of these calls crosses the device or the network. One call, to the model, leaves through every frame, and the device and network layers see it only as a connection or a domain. Neo Node sits inside the tool and decides each call before it runs.

Eight tools you already run. Eight places agents slip past.

  • Agentless SaaS and Cloud Tools

    AI apps and agents, through each account’s API

    SEES
    Sanctioned SaaS AI apps, cloud accounts and their permissions, through each one's API.
    STOPS
    At the API, and after the fact. Local agents, MCP servers, and extensions never reach it, and it changes a setting only once the action has happened.
    WITH NEO
    Neo connects to the same SaaS and cloud accounts, then covers the laptop no API reaches. One product, one policy, and decisions before the action runs.

  • Proxies and SSE

    Traffic between the device and the cloud

    SEES
    The domain and the device user on routed traffic, with content where TLS inspection is on.
    STOPS
    At the laptop’s edge. A file read or an rm -rf makes no traffic, and off the routed path it sees nothing.
    WITH NEO
    SSE keeps web, SaaS, and AI DLP policy on the network path. Neo decides each agent action on the laptop, in SaaS apps, and in the cloud, routed or not, and data on the laptop stays there.

  • MCP Gateways

    The MCP calls routed through them

    SEES
    Every MCP call routed through it: the server, the tool and its arguments.
    STOPS
    At its route. An MCP server on the laptop, and the agent’s built-in file and shell tools, never take the hop.
    WITH NEO
    You don't need a separate MCP gateway. Neo reads every MCP server, local or remote, with its grants, and decides each call before it runs.

  • Identity Providers

    Who signs in, and to which apps

    SEES
    Every sign-in that runs through it, and the OAuth grants each app holds.
    STOPS
    At the sign-in. The file read, the tool call and the push never pass back through it.
    WITH NEO
    Neo takes people and groups from it, so each record names the person, and whether an agent acted for them.

  • Prompt or API Layer

    What is sent to the model

    SEES
    Prompts, replies and tool requests, on traffic routed through it.
    STOPS
    At the model boundary. It sees the agent ask for a call, not what happens on the laptop after.
    WITH NEO
    You don't need a separate prompt layer. Neo decides on the action itself, with the agent, the session, and the content in hand.

  • MDM

    Installed apps and device settings

    SEES
    Installed apps, their versions and the assigned user.
    STOPS
    At the edge of the app. It lists Claude Code, not the MCP servers, skills and plugins loaded inside it.
    WITH NEO
    MDM deploys Neo Node. Neo Node controls what runs inside the apps.

  • EDR / XDR

    Processes, files and connections on the device

    SEES
    Every process, file open and connection, on or off the network.
    STOPS
    At agent context. A command the developer typed and the same command an agent issued look the same.
    WITH NEO
    Neo works inside the application, where the agent acts, and decides each action before it runs.

  • AI-Only Tools

    AI use, from the browser or the network

    SEES
    AI app usage and some agent activity, from wherever it sits: the browser or the network path.
    STOPS
    At AI. Binaries, packages and extensions without AI features need a separate allowlisting product.
    WITH NEO
    One Neo Node and one policy cover all software on the endpoint, AI or not, with default deny as an option.

Comparison reflects typical deployments. Capabilities vary by vendor and configuration.

One agent action. What each layer sees.

An engineer asks Claude Code to open a pull request. It reads a file and calls a GitHub MCP server on the laptop. Each row is a layer of your stack; each column, one part of that action.

Seen as what it isSeen only as a domain, process or textNot seen

LAYER WHAT IT SEES PersonAgentFileMCPServer Outcomesof 11
OVER THE NETWORK
BETWEEN AGENT AND MODEL Prompt or API Layer What is said to the model, including a file’s text after it is read. 0
ON THE NETWORK PATH Proxies and SSE The domains contacted, and the device user. 0
ON ROUTED MCP TRAFFIC MCP Gateways MCP calls routed through it. 0
INSIDE THE TOOL
NEO NODE AND CONNECTOR Neo Every part, as what it is and how the parts connect. 10
ON THE DEVICE
THE INSTALLED APP LIST MDM Which apps are installed, and for whom. 0
THE PROCESS VIEW EDR / XDR Processes, file opens and connections. 2

Outcomes count only those delivered in full.

Uncover every AI agent running in your company.

A live session with a security architect, on your questions, across endpoints, SaaS, and cloud. No canned demo.

Get a Demo See the Console

Get a Demo

See what Neo finds across your endpoints, SaaS, and cloud.

No canned demo. Visibility, threat control, cost, or something else entirely: our security architects map Neo to what matters to you.

Prefer email? info@neo.ai

By submitting, you agree to receive marketing communications from Neo; unsubscribe any time. See our Privacy Policy. Protected by reCAPTCHA; Google's Privacy Policy and Terms apply.

  • < 5 min First scan to full inventory
  • 25MB Neo Node, no kernel module, no reboot
  • < 15 min Full deployment with IdP and SOC