RelayRailRelayRail
Back to blog
AnnouncementDecember 15, 20245 min read

Introducing RelayRail: Human-in-the-Loop for AI Agents

Today we are excited to announce RelayRail, a new way for AI agents to communicate with humans via SMS and email.

We're excited to announce RelayRail, the human-in-the-loop primitive for AI agents. RelayRail gives your agents a phone number with a single MCP server, enabling them to orchestrate approvals, notifications, and commands via SMS and Email.

The Problem

AI agents are becoming increasingly autonomous. They can write code, deploy applications, manage infrastructure, and execute complex multi-step workflows. But with great autonomy comes great risk.

What happens when an agent is about to deploy untested code to production? What if it's about to delete files it shouldn't? Or send emails on your behalf without your knowledge?

Until now, the options have been limited: either give agents full autonomy (risky) or require constant supervision (impractical). Most developers end up with a fragile middle ground—agents that pause and wait for human input through whatever interface they happen to be running in.

The Solution

RelayRail provides a clean, simple API for agents to request human approval, send notifications, and receive commands—all through SMS and email. No matter where your agent is running, it can reach you.

Here's how it works:

  1. Install the MCP server: Add RelayRail to your Claude Desktop, Cursor, Cline, or any MCP-compatible agent in seconds.
  2. Agents request approval: When your agent needs permission for a risky action, it calls the request_approval tool. You get an SMS or email.
  3. You respond: Reply "YES" or "NO" via text, or click a button in the email. The agent receives your decision and proceeds accordingly.

Key Features

  • MCP Native: Works with Claude Desktop, Cursor, Cline, and any MCP-compatible agent out of the box.
  • Multi-Channel Routing: SMS for urgent approvals, email for less time-sensitive notifications. Let users choose their preference.
  • Context-Aware Response Pages: For complex decisions, we generate web pages with full context so users can make informed choices.
  • Enterprise Ready: API key authentication, usage tracking, and audit logs for compliance.

Getting Started

Getting started takes less than a minute:

npx @relayrail/server start

Add your API key, configure your phone number in the dashboard, and you're ready to go. Check out our Quick Start guide for detailed instructions.

What's Next

This is just the beginning. On our roadmap:

  • Voice call support for critical alerts
  • Team-based approval workflows (require 2 of 3 approvers)
  • Slack and Discord integrations
  • Webhook callbacks for programmatic response handling

We believe human oversight shouldn't be an afterthought—it should be a first-class primitive that agents can rely on. RelayRail makes that possible.

Ready to give your agents a phone number? Sign up for free and start building safer AI agents today.