withHuman

Set up human approval for your AI agents by asking Claude. withHuman checks your agents' tool calls against rules that you write. You can auto approve routine calls, refuse unsafe ones and asks a person about the rest. withHuman is a single centralized location for these rule across all your agents. With this connector you can: Write approval rules: "Approve refunds up to $50 and send larger ones to Finance." Set up escalation paths that say who is notified, in what order and outside office hours. Manage the agents that send requests, and their instances and credentials. Create teams and set how each one is paged. Test a sample request against your rules before they go live. Search the audit log and read how any request was decided. This connector manages withHuman. It does not put Claude's own tool calls behind approval. To do that, connect your agent with the withHuman CLI.