Overview
This article explains how Netskope counts AI Assets for billing purposes under the AI Command Center (AICC) subscription.
Definitions
AI Application – An AI Application is any web application identified as belonging to the Generative AI, Coding AI, or any other AI category by the Netskope App Catalog.
MCP Server – An MCP Server is any software application, service, or endpoint that communicates or processes data using the Model Context Protocol (MCP) over HTTP.
Agent: An autonomous software system driven by a large language model (LLM) that uses tools to achieve higher-level goals without constant step-by-step human control.
In AICC, the following AI assets are counted as Agents:
-
Agents built on agentic frameworks (example, LangChain)
-
AI browser extensions (example, Claude in Chrome)
-
AI assistants embedded in IDEs (example,Claude Code extension for VS Code)
-
AI desktop applications (example, Claude Desktop, ChatGPT Desktop)
AI Identity – An AI Identity is any user account or credentials used to access any AI Asset set forth in the AI Command Center documentation. AI Identities include any of the following:
- Corporate login credentials
- Personal accounts
- API tokens
Counting Methodology
AI Applications
Each entry in the Netskope App Catalog counts as one AI Asset, regardless of how many users or devices access it. If an AI Application is blocked for all users then it will not be counted as an AI Asset.
| Example | Count |
|---|---|
| ChatGPT.com accessed by 2,000 users on 2,000 devices | 1 AI Asset |
| ChatGPT.com and chatgptbrasil.com.br both in use | 2 AI Assets |
The second example counts as 2 AI Assets because ChatGPT.com and chatgptbrasil.com.br are listed as separate applications in the Netskope App Catalog.
MCP Servers
How MCP servers are counted depends on where they are installed.
Cloud/Web
Cloud/Web MCP servers are hosted remotely, in the cloud or on the web, and accessed over the network.
Cloud/Web MCP Server counting is based on the Fully Qualified Domain Name (FQDN). Each unique FQDN counts as one AI Asset. If a MCP Server is blocked for all users then it will not be counted as an AI Asset.
Example 1: Multiple MCP Servers on an Internal Network
An organization runs the following MCP servers on its internal network:
| FQDN | Count |
|---|---|
| mcp.prod.internal.netskope.com | 1 AI Asset |
| mcp.dev.internal.netskope.com | 1 AI Asset |
| mcp-backup.prod.internal.netskope.com | 1 AI Asset |
| Total | 3 AI Assets |
Each distinct FQDN counts as one AI Asset.
Example 2: Publicly Hosted MCP Servers, for example, Zapier.
The following URLs all share a single FQDN (actions.zapier.com):
https://actions.zapier.com/mcp/github https://actions.zapier.com/mcp/slack https://actions.zapier.com/mcp/<user_or_app_key>/sse
Counted as: 1 AI Asset. All paths under actions.zapier.com resolve to a single FQDN.
Example 3: Claude (Anthropic)
The following URLs have different subdomains and appear as separate rows on the MCP Server page:
https://microsoft365.mcp.claude.com/mcp https://pubmed.mcp.claude.com/mcp
Counted as: 2 AI Assets. Each subdomain appears as a distinct entry in the MCP Server page.
Local
Local MCP servers are installed and run on the endpoint. Each installed MCP server is counted as one AI asset.
-
Scenario 1: A Filesystem MCP server installed on two endpoints counts as two AI assets, regardless of whether the versions are the same or different.
-
Scenario 2: Two MCP servers installed on the same endpoint count as two AI assets.
Models
Each installed instance of a model is counted as one AI asset, whether on the same endpoint or across endpoints. An instance is a distinct model variant (example, llama3.2:3b and llama3.2:1b are separate models) on an endpoint.
-
Scenario 1: Different variants on different endpoints
llama3.2:3bon Endpoint-1 andllama3.2:1bon Endpoint-2 count as two AI assets. -
Scenario 2: Same model on multiple endpoints
llama3.2:3binstalled on two endpoints counts as two AI assets. -
Scenario 3: Different models on one endpoint
llama3.2andqwen2.5installed on Device-1 count as two AI assets. -
Scenario 4: One model accessed by multiple users
A model installed on one endpoint and accessed by two users from two other endpoints counts as one AI asset. Only the installation is counted.
Agents
Applies to: AI browser extensions, AI editor extensions, AI desktop applications, and agents built on agentic frameworks.
Each instance of an agent is counted as one AI asset. An instance is a running agent process, an installed extension, or an installed desktop application.
-
Scenario 1: Multiple agents on one endpoint
Two agent processes on one endpoint count as two AI assets of type Agent.
-
Scenario 2: Multiple agents on multiple endpoints
Two agent processes each of two endpoints count as four AI assets of type Agent.
-
Scenario 3: Same extension in multiple browsers
An AI browser extension installed in Chrome, Firefox, and Safari on one endpoint counts as three AI assets of type Agent.
-
Scenario 4: Same extension on multiple endpoints
An AI browser extension installed in Chrome on three endpoints counts as three AI assets of type Agent.
AI Identities
Each unique account or credential used to access any of the entities set forth in the AI Command Center documentation is counted as one AI Asset. Multiple identities belonging to the same employee each count separately.
Example: John Doe is a single employee with the following identities:
| Identity | Type | Counted |
|---|---|---|
| jdoe@netskope.com | Corporate SSO | Yes |
| john@gmail.com | Personal Google Workspace | Yes |
| johndoe@gmail.com | Client-specific Gmail | Yes |
| doe_test_user | Test account, no usage | No |
| api_key_prod_456 | Service API key | Yes |