Gong (Composio MCP) — agentic threat model

Not certain from the listing — The agent relies on external LLMs via the MCP host. The primary threat is indirect prompt injection where malicious instructions embedded in Gong call transcripts hijack the model's behavior during summarization.

The agent acts as a data retrieval pipeline for sensitive Gong transcripts, analytics, and deal insights. Threats include unauthorized data exfiltration of customer PII and proprietary sales strategies through compromised tool calls.

Orchestrated via Composio MCP. The primary threat is tool misuse, where an attacker manipulates the agent into executing unauthorized API calls (e.g., fetching transcripts the user shouldn't have access to) due to loose framework-level validation.

Composio manages the hosting and execution environment of the MCP server. Security relies heavily on the isolation of the Composio environment and the secure storage of the managed Gong access keys.

Not certain from the listing — There is no explicit mention of real-time guardrails, transcript sanitization, or logging of tool executions to detect anomalous data access patterns or injection attempts.

Authentication is handled via Composio's managed access keys. A key risk is privilege misalignment, where the agent inherits broad API permissions under the connected key, bypassing granular user-level access controls.

As an MCP tool, this agent can be composed with other agents in a multi-agent workflow. This introduces cascading risks where a compromised downstream agent could request and exfiltrate Gong's sensitive revenue data.