How to add human approval to a Claude agent with MCP in 5 minutes
Step-by-step: connect the redphonr MCP server to Claude Code, Claude Desktop or Cursor, and make your agent ask you on Slack or SMS before risky actions.
Human-in-the-loop for AI agents
Human approvals for AI agents, over MCP
redphonr pauses an AI agent before a risky action, asks a person on Slack, SMS, WhatsApp, Telegram, Discord, email or a phone call, and resumes the agent with their answer.
Agents call request_approval, ask_human, collect_input or verify_output from any MCP client: Claude, Cursor, the OpenAI Agents SDK, LangChain, CrewAI and more.
Slack, SMS, WhatsApp, Telegram, Discord, email and voice calls. On a call, press 1 to approve or 2 to deny.
If nobody answers, redphonr moves down the ladder and on to whoever is on call, then their backup. Weekly rotations with overrides.
Requests deny on timeout, one answer closes the request on every channel, and every decision is logged.
redphonr is a human-in-the-loop service for AI agents. It runs as an MCP server: an agent calls a tool such as request_approval, pauses, and resumes with the person’s decision as structured data.
Any MCP-capable agent, including Claude Desktop, Claude Code, Cursor, Windsurf, the OpenAI Agents SDK, LangChain, CrewAI, Mastra and AutoGen. Agents without MCP can use the HTTP API.
redphonr is free to start and does not need a credit card.
Step-by-step: connect the redphonr MCP server to Claude Code, Claude Desktop or Cursor, and make your agent ask you on Slack or SMS before risky actions.
We compared 10 ways to add human approval to AI agents: managed services, workflow tools and framework features. Channels, MCP support and when to use each.
From MCP-native escalation to PagerDuty, Twilio and Slack webhooks: 10 ways an AI agent can reach a person, ranked by how well they get an answer back.