AI agents that use the 101domain MCP Server depend on large language models (LLMs) to interpret natural language prompts. This means they may occasionally misunderstand your request or act on the wrong domain if instructions are unclear or too broad.
To reduce this risk, always design your integration with ‘human in the loop’ guardrails, especially when the agent can make changes to your account.
To implement guardrails:
If your AI application has write permissions (for example, updating DNS records or web forwarding), we strongly recommend that you:
-
Require explicit human confirmation before any high‑impact change is applied.
-
Clearly instruct the AI in your system prompt. For example:
'Before adding, changing, or removing web forwarding, you must ask the user to confirm the exact domain and destination URL.' -
Use API scopes, along with your AI agent’s tool authorization and selection rules, to follow a least‑privilege approach so the agent can only modify what it truly needs to.
These safeguards help prevent unintended changes when the model misinterprets a prompt or guesses at missing details.