Use cases

Two jobs, same handoff: approve the thing the agent should not do alone, and catch every agent's question in one list.

An account is one person, signed in with an email address or an Apple ID. Every agent that person connects calls the same server. The agent inbox page is the longer version of that idea. Install for Claude, ChatGPT, Cursor, and Codex is the human-in-the-loop MCP hub.

• Handoffs from more than one agent

  Claude, ChatGPT, Cursor, Codex, and custom agents call one MCP server. The tasks that need you land in one Invoke Human inbox.

• Approvals that survive closing the laptop

  How to send an agent approval through Invoke Human so the yes or no is a handoff with context, not a vague can I proceed.