mcp-builder

Warn

Audited by Gen Agent Trust Hub on Feb 22, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • COMMAND_EXECUTION (MEDIUM): The script provides a generic interface for executing local processes through the MCP stdio transport.
  • Evidence: In scripts/connections.py, the MCPConnectionStdio class accepts a command and args list, which are passed directly to mcp.client.stdio.stdio_client to spawn a subprocess.
  • Risk: This mechanism can be exploited to perform arbitrary command execution on the host system if the agent is tricked into passing malicious strings to the create_connection factory.
  • EXTERNAL_DOWNLOADS (LOW): The skill implements remote connection capabilities via HTTP and Server-Sent Events (SSE).
  • Evidence: MCPConnectionSSE and MCPConnectionHTTP classes in scripts/connections.py take a url parameter to establish outbound connections.
  • Risk: While these are standard transport methods for MCP, they provide a vector for data exfiltration or communication with attacker-controlled command-and-control (C2) servers.
  • INDIRECT_PROMPT_INJECTION (LOW): The connection utility acts as a high-privilege tool without input sanitization.
  • Ingestion points: Arguments passed to create_connection in scripts/connections.py.
  • Boundary markers: None present; the skill assumes the caller provides safe commands and URLs.
  • Capability inventory: Subprocess execution (stdio transport) and network operations (HTTP/SSE transport).
  • Sanitization: None; the code relies on the underlying mcp library for transport but does not validate the safety of the commands or endpoints themselves.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 22, 2026, 09:27 AM