CopyMarkdown

Security checkA97/100

We read every tool's name, description and inputs the way the model will, looking for instructions hidden from you. A clean result means no warning signs, not a guarantee. Scanned Nov 11, 2026.

• Hidden or reordered text
• Overriding the model's instructions
• Hiding actions from the user
• Reading keys or credentials
• Fake system tags
• Steering other servers' tools
• Pulling in the conversation
• Smuggling parameters
• Demanding to be used for everything
• Steering away from web search
• Download-and-run commands
• Encoded blobs
• Invisible characters
• Credentials as tool argumentsLow4 found

  Takes a password, API key or token as an argument, so the secret passes through the model's context and chat logs.

  • url_to_markdown · input.api_key.name
    api_key
  • generate_content_md · input.api_key.name
    api_key
  • generate_design_md · input.api_key.name
    api_key
  • generate_links_md · input.api_key.name
    api_key

  Fix: Read credentials from the server's own configuration (environment, OAuth) instead of a tool argument.

• Very long descriptions
• Read and write hintsNote1 found

  No readOnlyHint / destructiveHint annotations, so clients can't tell a read from a write without asking.

  • server · annotations
    4 of 4 tools: url_to_markdown, generate_content_md, generate_design_md, generate_links_md

  Fix: Add annotations to each tool, e.g. { readOnlyHint: true } for tools that only read.