Operating Rules
AI assistants need valid Workspace tokens and explicit context to operate safely. Ensure the assistant adheres to these rules:- Authenticate First: Every MCP tool call requires a valid Workspace token. Ensure your configuration has it set either via header or URL parameters.
- Resolve IDs Before Actions: Adding tags, starting flows, updating fields, or sending messages require internal IDs. Ask the assistant to verify target records and retrieve these IDs first.
- Review Bulk Actions: For broadcasts, bulk field updates, or mass messaging, require the AI assistant to list the target contacts and show the list before execution.
Human-in-the-Loop Approval
Human-in-the-loop means the AI assistant drafts and prepares an action, then pauses and waits for your confirmation before changing any data.Approval Rules to Instruct Your AI:
- Show target Contacts before any bulk action.
- Draft message content before sending.
- Resolve and display all target IDs before running actions.
- Test on one Contact before applying the action to a larger list.
- Report execution results once finished.