Our AI agents help run a platform with 70,000+ monthly users. The most important decision was not which model to use – it was what the agents are allowed to touch.
At Best-AI.org, OpenClaw and Hermes agents publish news and blog posts, curate a catalog of 3,900+ human-reviewed AI tools and handle CRM tasks. They do it through our own MCP servers – not with free access to the platform.
1. Tools, not access
Every capability is a defined MCP tool. A new capability means a new tool – a deliberate decision, not a side effect.
Our internal MCP servers cover platform administration: publishing news and blog posts, reviewing and curating tools, and the CRM for sign-ups and tool submissions.
2. Least privilege by default
Agents act through dedicated Google service accounts and Git access with scoped permissions. An agent gets the rights its tools need, nothing more.
3. Rules live in files, not in prompts
A hierarchy of rule files – general rules in instruction files in the agent root, task rules in a rule-file hierarchy in Google Drive – gives context and boundaries that people can read and change.
The admin side of Best-AI.org: agents work through internal MCP servers, bounded by rule files and scoped access.
The result: fast agents that stay under control. And humans stay in the loop for the catalog: every listed tool is human-reviewed.