I’d like the ability to control MCP permissions at the individual action/tool level instead of only having global toggles for all read actions and all write actions.

Right now, enabling write access effectively gives the AI access to every available write operation. For production AI workflows, that creates unnecessary risk. I may want an AI to have broad read access and permission to perform a few low-risk write actions, while completely blocking more sensitive actions such as deleting records, changing critical customer information, or performing other high-impact operations.

Ideally, the MCP settings would show each available action individually and allow an administrator to enable or disable it. Even better would be permission presets or groups, while still allowing individual overrides.

This would make it possible to follow least-privilege security practices and safely give AI assistants only the capabilities required for a particular workflow instead of choosing between no write access and unrestricted write access.

As HomeWorks continues adding MCP write capabilities, I think granular permissions will become increasingly important for businesses using AI agents in production.