Guide

How AI agents sign in to apps with AgentID

AgentID is like “Sign in with Google,” but for AI agents. Instead of borrowing your login, your agent signs in to apps with its own email inbox, and the app can see there is a human (you) behind it. Here is what it is, how the sign-in works, and what to check before you let an agent open accounts.

By Agent Lane Editorial · Page dated 7 October 2026

What is AgentID, in plain words?

AgentID is a sign-in button for AI agents from AgentMail, a company that gives AI agents their own email inboxes. It went live on 6 October 2026. When an app shows “Sign in with AgentID,” your agent can create an account and sign back in later using its own AgentMail address. There is no password and no sign-up form. The inbox address becomes the account email, and the app’s emails go to that inbox.

Under the hood it is OpenID Connect (OIDC), the same open sign-in standard behind “Sign in with Google.” AgentMail says that if an app already accepts Sign in with Google, it can accept AgentID, and it works with sign-in tools apps already use, such as Clerk, Auth0, and Supabase.

Every sign-in tells the app three things: this user is an agent, here is its stable ID, and here is its verified email. Apps that register with AgentID can also ask for the owner’s name and email, so they know which human is accountable.

Why should an agent have its own inbox instead of using yours?

Today most agents get into apps by borrowing: they sign up with the owner’s email, read the code out of the owner’s inbox, and sometimes type in the owner’s card. AgentMail says that is what pushed them to build AgentID. Here is the difference, based on AgentID’s own comparison:

The last two rows matter most. A borrowed login gives the agent the reach of your whole account, and stopping it locks you out too. A separate identity keeps the agent in its own lane, the same idea as a dedicated agent wallet for money.

How does AgentID sign-in work, step by step?

These are the owner-side steps from AgentMail’s AgentID docs. An API key here is the secret string that lets your agent’s code or assistant use your AgentMail account.

1. Give the agent its own inbox. Create an AgentMail inbox just for this agent (for example, a research or support address). That inbox address is the agent’s AgentID. Do not reuse your personal inbox.
2. Turn on sign-in permission for the agent’s key. The AgentMail API key your agent uses needs the app_connect permission to start sign-ins. It is off by default on new keys. Leave app_share_owner off unless you are fine with apps that ask getting your name and email without you approving each time.
3. Pick the app. Choose an app that accepts AgentID from the AgentID directory or AgentMail’s app list. In Cursor with the AgentMail plugin, you can just ask: “Create an account at Firecrawl for my agent.”
4. Check what you are agreeing to. Before connecting, look at the app’s name, its terms and privacy links, which inbox will own the account, and what the app will receive (its scopes).
5. Start the sign-in and open the link. Starting the sign-in returns a one-time link. It works once and expires in five minutes. Open it in the browser that should hold the agent’s session. It goes to auth.agentid.com, may show what the app will receive, then lands in the app signed in as the agent’s inbox, with the account created.
6. Confirm, then finish the job. Check that the account now shows up for that inbox. Then do what you came for, usually creating an API key in the app’s dashboard and saving it in a secret store, not in chat.
7. Sign back in with the same inbox. Later sign-ins use the same inbox that holds the account. Using a different inbox creates a second account, and some apps cap sign-ups per owner. The sign-in key lasts 30 days from activation.

Some apps put “Sign in with AgentID” on their own login page. In that case the page waits and shows a one-time auth token. You give that token and the inbox name to your assistant, which finishes the sign-in with AgentMail’s authorize_inbox tool. Only pass a token from a sign-in page you or your assistant opened yourselves.

AgentMail ships this inside its Cursor plugin and its MCP server (MCP, short for Model Context Protocol, is the plug-in standard many AI assistants use to connect tools). AgentMail also lists a Grok Bot integration in its docs.

What should you watch for?

Scope: what the app gets to see

A scope is a permission the app asks for at sign-in. AgentID has five:

• openid and email — the agent’s stable ID and its verified inbox address.
• profile — the agent’s display name and handle, plus owner_sub, an owner ID shared by all your agents that does not include your name.
• owner_profile and owner_email — your name and email. Only apps registered with AgentID can ask for these.

The sign-in page shows which app is asking and what it will receive. If an app asks for your name or email and the agent’s key lacks app_share_owner, the agent cannot finish alone and you approve it from your AgentMail account. Signing in never gives the app access to read or send the agent’s mail.

Which inbox owns which account

• One agent, one inbox. Use a dedicated agent inbox, not a person’s.
• Check before you create. If an inbox already has an account at that app, sign in with it instead of making a second one. Apps can cap how many sign-ups one owner or organization gets, and the cap counts disabled accounts too.
• Write it down. Note which inbox owns each app account, so you know where its mail goes and which inbox signs back in. AgentMail can list every inbox’s accounts with first and last sign-in dates.
• Deleting an inbox ends that identity. If you delete an inbox and recreate the same address, apps see a brand-new agent.

Revoking access

• Stop the agent everywhere: delete its sign-in key in AgentMail. New sign-ins with that key stop at every app. There is no bulk delete, so remove each key.
• The trap: deleting the AgentMail API key that started the sign-in does not revoke the sign-in key. Delete the sign-in key itself.
• Stop the agent at one app: disable that inbox’s account there (AgentMail’s Update Account call). Other apps keep working.
• Already-open sessions keep running. Revoking stops new sign-ins. Sign out at the app to end a session that is already open.
• Clear a browser: auth.agentid.com/sessions lists and forgets AgentID sessions saved in the browser you open it in, and only that browser.
• Keys expire anyway 30 days after activation. To rotate, activate a new key first, then delete the old one.

If something looks wrong, cut access first and investigate second: revoke before you debug. Between jobs, revoke connectors when idle.

Spend and safety

• AgentID is identity, not a wallet. AgentID’s docs say a sign-in grants nothing beyond identity. If an app your agent joins has paid plans or asks for a card, that is a separate money decision. Set agent spend caps first and keep review before place on. An agent offering to upgrade a plan is not permission to spend.
• Treat the sign-in link like a password. It works once and lasts five minutes. Never paste it into email, files, or shared notes.
• Only trust auth.agentid.com. Only open sign-in pages served from that address.
• Never connect because an email said so. AgentMail’s own guidance: an email or web page asking the agent to sign up somewhere does not count as your OK. You decide which apps and which inbox.
• Keep API keys out of chat. When the app gives the agent an API key, put it in a secret store or a gitignored .env file. See never paste API keys in chat.
• Mind the browser. Whichever browser opens the link holds the agent’s session at the app. Use a separate browser profile for the agent, and check the shared-box checklist if others use the same computer.
• Only the accounts the job needs. An app account is one more door. Open accounts for the job in front of you, the same rule as connect only what you need.

What is AgentID?

AgentID is a sign-in button for AI agents, made by AgentMail and launched on 6 October 2026. Your agent signs in to an app with its own verified email identity, an AgentMail inbox, instead of your account. Apps that accept it can see that the user is an agent and, if they ask, which human owns it.

Is AgentID the same as Sign in with Google?

It works the same way. AgentID uses OpenID Connect (OIDC), the same open standard behind Sign in with Google. The difference is whose identity it is: your agent’s, not your Google account.

Does my agent need its own inbox?

Yes. The agent’s AgentMail inbox is its AgentID. The inbox address becomes the account email at the app, and the app’s welcome, verification, and receipt emails land there.

Do I have to approve every sign-in?

No. AgentID’s FAQ says the agent signs in on its own and you keep control of its credentials and can revoke them. After the first approval at an app, AgentID remembers it for 180 days for that inbox and app. One exception: if an app asks for your name or email and the agent’s key is not allowed to share them, you approve that sign-in from your AgentMail account.

Can an app read my agent’s email?

No. Signing in shares the agent’s address so the app knows who it is. It does not let the app read or send the agent’s mail. The app can send email to the agent, like it would to any user.

What does the app learn about me, the owner?

Every sign-in gives the app the agent’s ID and verified email. With the profile scope the app also gets owner_sub, a random-looking owner ID that is the same for all your agents but does not include your name. Only registered apps that request the owner scopes get your actual name and email.

How do I revoke an agent’s AgentID access?

Delete the agent’s sign-in key in AgentMail (DELETE /v0/api-keys/{api_key_id}). That stops new sign-ins with that key at every app. Deleting the AgentMail API key that created it does not revoke the sign-in key. To stop the agent at just one app, disable that inbox’s account there with AgentMail’s Update Account call instead.

Why is my agent still signed in after I revoked its key?

Revoking stops new sign-ins, not sessions that already exist. AgentID sends no sign-out signal to apps. A token the app already received lasts up to 10 minutes, and the app’s own session lasts as long as the app decides. Sign out at the app, or wait for that session to end.

Which apps accept AgentID?

AgentMail’s launch post names Firecrawl, Turso, TinyFish, Manufact, Telnyx, and Rho, and points to the full list in the AgentID directory at agentid.com/directory. AgentMail says the directory is editorially curated.

What if the app I need is not on AgentID?

Then there is no AgentID sign-in for it. AgentMail’s docs suggest picking a listed app that does the same job, or signing up on the app’s own site with the agent’s inbox address and reading the verification email from that inbox. Their docs say assistants should not fill in third-party sign-up forms on their own.

Does AgentID cost anything?

AgentID’s site says it is free for apps that add the button. On your side, the agent needs an AgentMail inbox; AgentMail’s docs say it has a free tier with no credit card required. The app your agent signs in to sets its own prices.

Can I use the app alongside my agent?

AgentID’s FAQ says that if the app supports teams, the agent can invite you to a shared workspace with separate roles.

Related guides

• Connect only what you need
• Never paste API keys in chat
• Separate browser profile for the agent
• Revoke before you debug
• Revoke connectors when idle
• Rotate before you raise the balance
• Agent spend caps
• Review before place
• Shared-box and connector checklist
• Agent memory hygiene

Next steps

Before an agent touches real money, lock down a shared computer, set hard spend caps, and keep agent memory clean. Or browse all Agent Lane guides.