Tutorial · MCP · Claude

Give your Claude agent an email inbox in 60 seconds

Agents keep bumping into email: a service wants to verify an address, a workflow needs a human's emailed answer, a legacy system only sends notifications over SMTP. This tutorial wires a real inbox into Claude with one MCP config entry — no domain, no DNS, no OAuth dance.

Step 1 — get an API key

Sign up free at mxu.io (no card) and create a key in the dashboard. Free accounts get one receive-only inbox with 200 inbound messages/month; mail expires after 48 hours.

Step 2 — add the MCP server

# Claude Code (CLI)
claude mcp add --transport http mxu \
  https://api.musicsupplies.com/functions/v1/mxu-mcp \
  --header "Authorization: Bearer mxu_live_YOUR_KEY"

Claude Desktop or any MCP client, in JSON:

{
  "mcpServers": {
    "mxu": {
      "type": "http",
      "url": "https://api.musicsupplies.com/functions/v1/mxu-mcp",
      "headers": { "Authorization": "Bearer mxu_live_YOUR_KEY" }
    }
  }
}

Step 3 — try it

Ask Claude:

"Create an inbox called scout, then wait for a message and
 tell me what it says."

Claude calls create_inbox(name: "scout") and starts wait_for_message. Send anything to scout@mxu.io from your own mail client — within seconds the agent has it, including extracted.otp, extracted.links, and extracted.verify_link parsed and ready.

Real patterns this unlocks

Verification flows: the agent signs up for a service with its own address, waits, and enters the extracted code — no human copy-pasting OTPs.
Human-in-the-loop by email: the agent emails a person a question (paid plans), blocks on the reply, and continues the task with the answer. Replies are threaded properly, so the human just sees a normal email conversation.
System notifications: point a cron job's MAILTO, a monitoring alert, or any "email notifications only" SaaS at the agent's address; add a webhook if you'd rather push than poll.

How it stays trustworthy

Extraction is deterministic (parsers, not a model), the free tier is receive-only so nobody can spam from it, and every send on paid plans is DKIM-signed and rate-guarded — which is why addresses at the shared domain keep arriving in inboxes instead of spam folders.