rudder-mcp-workflow

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The workflow uses an MCP client to call mcp.rudderstack.com/mcp tools that return workspace data such as live events and destination errors; those tool responses can include free-form text authored by outsiders (e.g., customer event payloads, error messages) which the agent ingests into its LLM context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill requires an active runtime connection to the MCP endpoint https://mcp.rudderstack.com/mcp (used by mcp-remote) which dynamically returns tool lists and responses that directly control the agent's available instructions and actions, so this external URL is a runtime dependency that controls prompts.