MCP Server Inventory
The MCP Servers tab under Discover AI > AI Tools lists the MCP servers and plugins that AI tools on your managed devices have configured or called. Use it to see which third-party systems your workforce has connected to AI tools.
MCP servers are reported by the macOS agent only. Older macOS agents and the Windows agent do not report them, so devices running those may still use MCP servers that do not appear here.
Where the data comes from
The macOS agent reads the following on each device:
| AI tool | What the agent reads |
|---|---|
| Claude Code | Claude Code's local session logs: the MCP servers and plugins available in a session, and each mcp__ tool call. |
| Claude Desktop | Claude Desktop's MCP log (~/Library/Logs/Claude/mcp.log): the server name on each tool call. |
| GitHub Copilot | The server names under mcpServers in ~/.copilot/mcp-config.json, and MCP tool calls recorded in VS Code Copilot Chat sessions. |
| Gemini CLI | The server names under mcpServers in ~/.gemini/settings.json. These show as configured; calls are not counted. |
From configuration files the agent reads only the server names. It does not read or send server commands, arguments, URLs, environment variables or credentials.
For each server, the agent sends:
- The server name.
- The type: an MCP server, or a Claude Code plugin.
- The AI tool it was used from.
- The number of tool calls in the reporting window. Zero means the server is configured but was not called.
- For Claude Code plugins, whether the plugin is connected or still needs sign-in.
It does not send tool call arguments or results.
The MCP Servers tab
| Column | Meaning |
|---|---|
| MCP Server | The server name, with a badge: MCP for an MCP server, Plugin for a Claude Code plugin. |
| Used From | The AI tools the server was configured in or called from. |
| Devices, Users | Distinct devices and people. |
| Tool Calls | Calls in the selected period. A server that is configured but was never called in the period shows that instead of a count. |
| Status | Connected or Needs sign-in, where known. |
| First Seen, Last Seen | When the server was first and last reported. |
Search by server name, and filter by the AI tool the server was used from. The tab also honours the time range and the OS and Department filters.
Who uses each server
Select a server to see where it is used: one row per user and device, with the AI tool it was configured in or called from and when it was last seen. Use it to find who has a risky server installed and on which laptop.
Export
Select Export CSV to download the list with the same filters applied. Use the export for security reviews and as a starting point for an allow-list.
Storage and retention
ANTS Platform stores each MCP report against the organization, the user and the device that sent it. MCP records follow your organization's data retention setting and are removed by the nightly retention purge with the rest of your Discover AI data.
Limits
- Discovery covers macOS devices. The Windows agent counts MCP tool calls from the Claude desktop app's log but does not report server names.
- MCP servers in CI/CD pipelines are not scanned.
- The tab is an inventory. Enforcing an allow-list of MCP servers is on the roadmap.
Switching it off
MCP data comes from the agent's log readers for Claude Code, the Claude desktop app, VS Code Copilot Chat and Gemini CLI. Switching off a tool's log reader in the collector settings stops its MCP reporting too, along with that tool's prompt, tool-call and token counts. See Collector settings.