using-the-mcp-server

Pass

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses a launcher script (scripts/mcp-launch.sh) documented to download a prebuilt binary from the tool's GitHub release if it is not found on the local path or via package managers like npx, uvx, or Homebrew.
  • [REMOTE_CODE_EXECUTION]: The launcher script is designed to execute the html-to-markdown binary immediately after resolving or downloading it. Executing binaries downloaded from external sources at runtime presents a significant security risk if the source repository or delivery channel is compromised.
  • [COMMAND_EXECUTION]: The skill documentation details the manual and automatic execution of shell commands, specifically html-to-markdown mcp, to initialize the MCP server.
  • [PROMPT_INJECTION]: The convert_html and extract_metadata tools ingest arbitrary HTML strings, creating a surface for indirect prompt injection where malicious instructions embedded in the HTML could influence the agent's behavior.
  • Ingestion points: The html parameter in both convert_html and extract_metadata tools.
  • Boundary markers: None identified in the documentation to isolate untrusted HTML input.
  • Capability inventory: The skill executes local binaries via the bundled launcher script.
  • Sanitization: No input sanitization or validation of the HTML content is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 22, 2026, 10:22 PM
Security Audit — agent-trust-hub — using-the-mcp-server