How to give Cursor IDE a 'Phone a Friend' lifeline using MCP

AI-assisted "vibe coding" with tools like Cursor and Claude Desktop is incredible, until it isn't. We've all been there: the agent gets stuck in a loop, suggesting the same broken fix over and over, or it confidently uses an API method that was deprecated three years ago.

When an agent is stuck, what if it could just ask a person for help?

With the Model Context Protocol (MCP), you can give your agent that lifeline. Here is how to add askingahuman.com to Cursor and Claude Desktop, so your agent can reach a real person when it needs one.

What is the Model Context Protocol (MCP)?

MCP is an open standard that lets AI models connect to external tools and data. Instead of relying only on its training data, an agent with MCP can read your files, query a database, or, in our case, reach a person.

Step 1: Get an API key

• Go to the Developers page and create a free API key. It's shown once, so save it.
• Free questions need nothing else. For paid live chats, add a card on the same page.

Step 2: Add it to Cursor

1. Open Cursor Settings, go to the MCP section and choose New MCP Server. This opens your mcp.json file.
2. Add the askingahuman server, with your key in place of YOUR_API_KEY:

{
  "mcpServers": {
    "askingahuman": {
      "command": "npx",
      "args": ["-y", "@askingahuman/mcp-server"],
      "env": { "ASKINGAHUMAN_API_KEY": "YOUR_API_KEY" }
    }
  }
}

Save the file. Cursor picks up the askingahuman tools on its own.

Step 3: Add it to Claude Desktop

Open Claude Desktop's settings, go to Developer → Edit Config, and add the same askingahuman entry to claude_desktop_config.json. Restart Claude Desktop.

How it works in practice

When your agent is stuck on a hard architectural decision or a strange bug, you can prompt it:

"I'm not sure how to fix this AWS Cognito routing issue. Find someone on askingahuman who knows AWS and ask them."

Every request goes to a person the agent chooses. It looks people up with list_humans or sees who is online with who_is_live. Then it can:

• Ask a free question with ask_human, to someone who takes them. The person answers when they're free, and the agent reads the reply with check_answer.
• Start a live chat with connect_human. The agent checks the person's price first with get_quote, and you approve the cost. Nothing is charged unless the person accepts.

You stay in control of spending: set ASKINGAHUMAN_MAX_SPEND_USD in the same env block to cap what one chat can cost.

It's the failsafe for autonomous agents. Stop the loops. Let your AI ask a human.