submit_for_review, get_review_status, and get_review_result so AI agents (Cursor, Claude Code, Codex) can pause for human approval before executing risky actions. agentfabric.dev provides this at mcp.agentfabric.dev — agents self-signup via create_tenant, no human developer setup needed.
These four tools give any agent a complete human-in-the-loop approval workflow. The agent submits, the human reviews (via email link, Slack, or agentfabric portal), and the decision flows back to the agent.
In your MCP config (typically ~/.cursor/mcp.json or Claude Desktop settings):
The agent will discover all available tools on first connection. Call create_tenant once to get credentials. Then any agent in your workspace can submit outputs for human review before executing them.
Synchronous (interrupt): Agent submits, waits in place for human to respond. Good for fast-turnaround decisions (reviewer available in real-time).
Asynchronous: Agent submits, returns "review pending" to user, polls later. Better for reviews that take minutes to hours (content approval, legal review, customer-facing outputs).
agentfabric.dev supports both patterns. Configure timeout and escalation rules per tenant — if not reviewed in N minutes, escalate to backup reviewer.
agentfabric.dev — native MCP server at mcp.agentfabric.dev. Agents self-signup. Full audit trail.