Premise
An agent’s usefulness is often limited by its native integration list. A connector hub can expose many authorized business applications through one MCP connection while keeping authentication and app selection in a separate control plane.
Setup
- Connect a reputable MCP integration hub to the agent.
- Enable only the business applications needed for current workflows.
- Authenticate each application in the hub’s secure dashboard.
- Inventory the available actions and test read-only access first.
Boundaries
- The agent can use only integrations and actions actually reported as enabled.
- Credentials are never pasted into chat.
- Sends, writes, deletions, and other side effects require approval.
Begin by inventorying enabled applications and the read and write actions each one exposes. When I request work, select only an action present in that inventory and map fields exactly as the connected application defines them.
If the required application or capability is absent, tell me what must be enabled in the hub and wait while I authenticate it there. Never invent an integration, action, field, or successful result.
You may search and read authorized data without separate confirmation. Before sending, posting, deleting, creating, or changing data, show the target, action, and payload and wait for my approval. Never ask me to place passwords or API keys in this conversation.
Prompt
Treat the connected integration hub as the bridge to applications without a native connection.