mcp-tool-discovery

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to ingest and act upon tool metadata (names and descriptions) fetched from a ServiceNow instance via the tool_search meta-tool. If this external content is controlled by an adversary, it could potentially contain instructions aimed at influencing the agent's behavior during tool selection and execution.
  • Ingestion points: Metadata returned by the ServiceNow MCP server via tool_search (SKILL.md).
  • Boundary markers: The instructions do not specify any delimiters or safety prompts to ignore instructions found within tool descriptions.
  • Capability inventory: The skill enables the tool_execute capability, which allows for data modification on the ServiceNow instance (SKILL.md).
  • Sanitization: No sanitization or verification of the external tool schema or description text is mentioned.
  • [COMMAND_EXECUTION]: The documentation provides configuration examples for launching the @serac-labs/servicenow-mcp package using npx. This is the official server implementation from the vendor (serac-labs) used to establish the MCP connection.
  • [PERSISTENCE]: The skill describes how the server persists enabled tool state to the local filesystem (e.g., ~/Library/Application Support/snow-code/ on macOS). This is standard application behavior for maintaining session state and does not constitute an unauthorized persistence mechanism.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 07:16 PM
Security Audit — agent-trust-hub — mcp-tool-discovery