SingleUseMail
Get the app Get an address
For your agents

Let your agent finish the signup, not read your mail.

Agents stall at verification screens because they have no inbox. Give each of yours a key of its own: it asks for an address, gets the code, and nothing more. You approve new sites, and you see every request.

Connect an agentWhich agents work
Works with anything that speaks MCP: Claude, ChatGPT connectors, Cursor, Codex, custom agents. Two minutes, no account.
NEW SITE

Claude wants an address for notion.so

First time this agent asks for this site. Codes and links only. Allowing "always" skips this next time for notion.so only.
Allow onceAlways for this siteDeny and pause
No answer in 10 minutes counts as no. The agent is told to stop and wait for you.
How it works

Four steps, then it just works

Name the agent and pick its rules.
In Your agents, press Connect an agent. "Codes and links only" and "Ask me before new sites" are on by default.
Paste its connection into the agent.
We show a URL that carries the agent's own key. Add it as an MCP server or connector where your agent lists tools. Do it once.
The agent asks for an address when it needs one.
It calls address_for("site"). Same site, same address, every time, so it never has to remember inbox IDs.
The code comes back as text.
It calls wait_for_code("site") and gets the six digits or the link, usually within seconds. The email itself stays with you.
What the agent sees
address_for("acme-app.com")
→ k7m3q2xa.h4@niirii.com

wait_for_code("acme-app.com")
→ code: 482913
  from: no-reply@acme-app.com
  subject: Your verification code
What the agent never seesYour key. Your other addresses. Other agents' addresses. The email body, unless you allow it for that agent. It cannot send, reply, forward or delete anything.
Which agents work

Anything that can add an MCP server or a connector

The connection is a standard MCP server over HTTPS. Where your agent lets you add a tool server, paste the URL from the Connect screen. Exact menu names change with each release, so the connect screen shows the current steps for the agent you pick.

ClaudeAdd as a custom connector (remote MCP) in Settings → Connectors, or as an MCP server in Claude Code and Claude Desktop config.
ChatGPT and CodexAdd as a connector or MCP server where developer-mode tools are configured. Codex CLI reads MCP servers from its config file.
Cursor, Windsurf, and IDE agentsAdd to the MCP servers list with the URL as a streamable HTTP server. No command, no environment variables.
Browser agentsAgents that drive this browser need no key: they can open /app and read the address list you see. Keep the tab signed in with your key.
Custom agents and scriptsCall the same tools over plain HTTP with the agent key as a Bearer token. Two endpoints, JSON in, JSON out. See the developer docs.
Something else?If it can make an HTTPS request, it can use an agent key. Tell us which agent and we will add the steps.
The connection, in one line
https://api.singleusemail.com/mcp/ak_<your agent's key>
The key in the URL is the agent's, never yours. Treat the URL like a password for that one agent. Remove the agent to void it.
Safety rules

What every agent can and cannot do

Codes and links only is on by default. The agent gets the verification code or the link from the newest email for that site, nothing else.
Ask me before new sites is on by default. The first request for a site waits for your approval: once, always for that site, or deny.
Deny pauses the agent. A denied agent stops getting anything until you resume it. Ten minutes without an answer is a deny.
Every request is recorded. Address asked, code returned, approval given or refused, with the time. The record lives with your key.
Never the key. Agent keys are derived from yours one way. Nothing an agent holds can recover your key or reach addresses it was not given.
Never sending. The whole service is receive-only. No agent can send, reply, forward, or delete email.
Never other agents' addresses. Each agent has its own address space. Two agents asking for the same site get different addresses.
Remove is instant. Removing an agent voids its key at once. Its addresses stay in your list so you can still read what arrives.
Agent allowances

Addresses are free. Codes are counted.

An agent can ask for as many addresses as it likes; nothing is stored until email arrives. What we count is codes an agent actually receives, across all your agents, per month.

Free
20 agent codes / month
  • Unlimited agents and addresses
  • Approvals and the record
  • Codes and links only, or full email per agent
Connect an agent
Plus
200 agent codes / month
  • Everything in Free
  • $9 per further 1,000 codes
  • 30-day emails, private domain, no sponsor slot
See Plus
Your product's agents
API per code, per tenant
  • Keys per tenant and per agent
  • Your own domains
  • Metered billing, webhooks
Developer docs
Questions

Before you connect one

Does the agent get my key?

No. Each agent gets its own key, made from yours in this browser. It can only ask for addresses and codes, and only within the rules you set. Removing the agent stops it instantly; your key and your other addresses are untouched.

Can an agent read the whole email?

Only if you turn off "Codes and links only" for that agent. Most agent tasks need the six digits or the link, nothing else. Turning it on later needs your explicit change; it never happens by itself.

What if an agent starts signing up for things I did not ask for?

With "Ask me before new sites" on, the agent waits for your approval the first time it asks for a site. Deny and the agent is paused. Every request, allowed or denied, is in the Record.

Does this work with browser agents that drive my browser?

Yes, without a key. An agent that controls this browser can open singleusemail.com/app and use the same address list you use. The browser extension makes that a one-click fill once it ships.

Is there a limit?

Free includes 20 agent codes a month across all your agents. Plus includes 200, then $9 per 1,000. Addresses are never counted; only codes an agent actually receives.

Can I use this for my product's agents?

That is the developer door. The same engine is an API and MCP server with per-tenant keys, your own domains and metered billing.

Connect an agentHow the key works