🚧
Special privileges required
The Agentic platform and its MCP server are only available to users with Google Threat Intelligence (Google TI) Enterprise or Enterprise Plus licenses.

The GTI Agentic MCP Server lets you connect any assistant or environment that speaks the Model Context Protocol (MCP) directly to the intelligence of Google Threat Intelligence (Google TI) and VirusTotal. Instead of switching to the web UI, you can drive natural-language threat hunting, IoC analysis, and multi-turn investigations from your own MCP-compatible client — IDEs, desktop assistants, or custom agents.

The server exposes the same Agentic capabilities available in the Agentic platform as a set of MCP tools, so your client can ask questions, continue investigations, and manage research sessions programmatically.

Table of contents

Requirements
Standard MCP configuration
Available tools
Usage examples

Requirements

To connect to the GTI Agentic MCP Server you need:

A VirusTotal account.
A VirusTotal API key. See How to get GTI API keys for how to locate or create one.

⚠️
Keep your API key private
Your API key authenticates every request as you and consumes your quota. Treat it like a password: do not commit it to source control, share it, or paste it into untrusted clients.

Standard MCP configuration

The GTI Agentic MCP Server is a remote MCP server. Most MCP clients accept a standard configuration block that points at the server URL and passes an authentication header. The two parameters you need are:

Server URL: https://www.virustotal.com/api/v3/agentspace/mcp/
Authentication header: x-apikey: <YOUR_VIRUSTOTAL_API_KEY>

Add the following block to your client's MCP configuration, replacing <YOUR_VIRUSTOTAL_API_KEY> with your own key:

{
  "mcpServers": {
    "gti-agentic": {
      "url": "https://www.virustotal.com/api/v3/agentspace/mcp/",
      "headers": {
        "x-apikey": "<YOUR_VIRUSTOTAL_API_KEY>"
      }
    }
  }
}

📘
Client-specific configuration
The exact location and format of the MCP configuration vary by client. Some clients read a mcpServers object like the one above, while others expose the URL and headers through a settings UI. Refer to your client's documentation for where to place the server URL and the x-apikey header.

Once the server is configured and your client reconnects, the gti-agentic tools listed below become available to the assistant.

Available tools

The server exposes 7 tools that map to the core Agentic investigation workflow:

Tool	Description
`ask_gti_agent`	Run analytical queries and threat hunting in natural language — IoCs, threat actors, malware families, and summaries. Starts a new investigation.
`continue_chat`	Continue an interactive, multi-turn investigation with follow-up questions in an existing session.
`list_sessions`	List your active and historical investigation sessions.
`get_session_history`	Retrieve the full history of events and messages for a session.
`delete_session`	Delete and clean up finished sessions.
`share_session`	Share an investigation session so a colleague can access it.
`copy_shared_session`	Copy (fork) a shared session to continue it on your own.

📘
Sessions carry investigation context
ask_gti_agent opens a session; continue_chat keeps the conversation grounded in what was already found. Use list_sessions and get_session_history to pick up an earlier investigation, and share_session / copy_shared_session to collaborate — the same session model behind the Agentic platform.

Usage examples

Once connected, you can prompt your assistant with requests such as:

IoC lookups — reputation, maliciousness, and known associations for hashes, domains, and IP addresses.
- "What is the current reputation and known threat-actor associations of the IP address xxx?"
Triage of recent samples — surface and prioritize recently seen samples with high detection.
- "Triage the most-detected samples related to hash xxx seen in the last 7 days."
Threat summaries — concise briefs by sector, campaign, or threat actor.
- "Draft an executive threat brief on the recent activity of threat actor xxx over the last 90 days."

Follow-up questions in the same session (via continue_chat) let you drill down — pivoting from an IoC to its related campaigns, or expanding a summary into a detailed report — while Agentic keeps the earlier findings and citations in context.