← All guides

Set Up Disposable Email for a Claude Agent

Claude can work a signup flow right up to the point where the service emails a code, and then it has nowhere to read it. Adding myagentinbox as an MCP server gives Claude an inbox of its own, plus the tools to check it and pull the code out.

Pick your Claude

• Claude Code — terminal CLI, best for autonomous coding tasks
• Claude Desktop — chat UI on Mac/Windows
• Claude Agent SDK — programmatic agents in TypeScript or Python

---

Claude Code

1Add the MCP server. From your project directory, run:

claude mcp add myagentinbox npx mcp-remote https://myagentinbox.com/mcp

That registers the server for the current project. Use --scope user to make it available across every project on your machine:

claude mcp add --scope user myagentinbox npx mcp-remote https://myagentinbox.com/mcp

2Verify it loaded. Start Claude Code and run /mcp. You should see myagentinbox listed with four tools.

3Use it. Just ask:

> sign me up for the example.com newsletter using a throwaway email,
  then read the welcome message and tell me what's in it

Claude will call create_inbox, use the returned address in the signup form, poll check_inbox, and read the message back to you.

---

Claude Desktop

1Open the config file.

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

2Add the server under mcpServers:

{
  "mcpServers": {
    "myagentinbox": {
      "command": "npx",
      "args": ["mcp-remote", "https://myagentinbox.com/mcp"]
    }
  }
}

3Restart Claude Desktop. Look for the tools icon in the chat input — myagentinbox should appear in the connected servers list.

---

Claude Agent SDK

The Claude Agent SDK lets you build long-running agents in TypeScript or Python. MCP servers are passed in via mcpServers.

TypeScript

import { query } from "@anthropic-ai/claude-agent-sdk";

const result = query({
  prompt: "Sign up for service X with a throwaway email and report back the verification code.",
  options: {
    mcpServers: {
      myagentinbox: {
        type: "http",
        url: "https://myagentinbox.com/mcp",
      },
    },
    allowedTools: [
      "mcp__myagentinbox__create_inbox",
      "mcp__myagentinbox__check_inbox",
      "mcp__myagentinbox__read_message",
    ],
  },
});

for await (const msg of result) {
  console.log(msg);
}

Python

from claude_agent_sdk import query, ClaudeAgentOptions

options = ClaudeAgentOptions(
    mcp_servers={
        "myagentinbox": {
            "type": "http",
            "url": "https://myagentinbox.com/mcp",
        },
    },
    allowed_tools=[
        "mcp__myagentinbox__create_inbox",
        "mcp__myagentinbox__check_inbox",
        "mcp__myagentinbox__read_message",
    ],
)

async for message in query(
    prompt="Sign up for service X and read the verification email.",
    options=options,
):
    print(message)

---

The four tools, in plain English

A worked example: verification code flow

Here's the typical loop your agent will run, narrated for clarity:

1. Call create_inbox → receive a7k2m9x4bp@myagentinbox.com.
2. Submit that address to the target signup form (via browser tool, fetch, or whatever the agent has).
3. Wait 5–10 seconds, then call check_inbox. If empty, wait and retry — most providers send within 30 seconds.
4. Once a message appears, call read_message with its id. Extract the code or magic link from the body.
5. Submit the code back to the target service. Done.

Limits to know

• Inbox creation: 1 per 5 minutes, and 10 per day, per IP
• API reads: 20 per minute per IP
• Inbox lifetime: 24 hours, then everything is deleted
• Max email size: 10 MB

What to use it for

• Account signups that require email verification
• One-time passwords and magic-link logins inside agent test runs
• Receiving notifications or webhook-style emails as part of a workflow
• Pulling attachments (PDF receipts, exported reports) into a downstream pipeline

What not to use it for

• Anything you need to keep — inboxes are gone after 24 hours.
• Long-lived account recovery email — pick a real mailbox.
• Sending email — myagentinbox only receives.

---

That's the whole setup. If something doesn't work, the landing page has a one-click "create inbox" button you can use to confirm the service is reachable from your network before debugging the MCP layer.

Questions or feedback: hello@myagentinbox.com.