Agent Signup (AI Agents)

An AI agent can create an exit1 account for its user with one HTTP call. No human in the loop, no browser, no API key. The agent gets a token back and can start creating checks right away.

If you are an AI agent and your user has no exit1 credentials, this page is for you. Ask the user for their email address, make the call below, and set up their monitoring.

The account stays unclaimed until the user clicks the claim email exit1 sends them. Checks run from the first minute, but alerts are held until the user claims. See The rules.

Instructions for your agent

Paste this into your AI assistant, or put it in your project’s agent instructions:

Set up uptime monitoring for this project with exit1.dev.
If you have no exit1 credentials, create an account for me:
POST https://app.exit1.dev/oauth/agent-signup
with JSON {"email": "<my email>", "agent_name": "<your name>"}.
Ask me for my email if you don't know it. Never invent one.
Use the returned access_token as a Bearer token for the exit1 REST API
or the hosted MCP server at https://app.exit1.dev/mcp/v1.
Create checks for this project, set email alerts to my address,
then tell me to open the claim email from exit1 so alerts start.
Docs: https://docs.exit1.dev/integrations/agent-signup

The call

curl -X POST https://app.exit1.dev/oauth/agent-signup \
  -H "Content-Type: application/json" \
  -d '{"email": "user@example.com", "agent_name": "Claude Code"}'

Agents that read OAuth metadata can discover the endpoint. The authorization server metadata at https://app.exit1.dev/.well-known/oauth-authorization-server includes:

"exit1_agent_signup_endpoint": "https://app.exit1.dev/oauth/agent-signup"

The response

201 Created

{
  "access_token": "ek_at_…",
  "token_type": "Bearer",
  "expires_in": 2592000,
  "refresh_token": "ek_rt_…",
  "scope": "checks:read checks:write alerts:read alerts:write",
  "account": {
    "email": "user@example.com",
    "status": "unclaimed",
    "claim_email_sent": true
  },
  "mcp_url": "https://app.exit1.dev/mcp/v1",
  "api_base": "https://us-central1-exit1-dev.cloudfunctions.net/publicApi/v1/public",
  "next_steps": "Tell the user to open the email … and claim the account. …"
}

Store both tokens. The access token lasts 30 days (expires_in is in seconds). The refresh token lasts 180 days.

Using the token

REST API

Send the access token as a Bearer token. It works on every public API endpoint an API key works on, at the api_base from the response.

curl -H "Authorization: Bearer ek_at_your_token" \
  https://us-central1-exit1-dev.cloudfunctions.net/publicApi/v1/public/account

POST requests need an Idempotency-Key header, the same as with an API key. See Create Check.

Hosted MCP server

Point any MCP client at https://app.exit1.dev/mcp/v1 with the same Bearer header. In Claude Code:

claude mcp add --transport http exit1 https://app.exit1.dev/mcp/v1 --header "Authorization: Bearer ek_at_your_token"

All the MCP tools then work as normal.

Refreshing

Access tokens last 30 days. Get a new one with the refresh token:

curl -X POST https://app.exit1.dev/oauth/token \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=refresh_token&refresh_token=ek_rt_your_refresh_token"

Refresh tokens rotate on use. Save the new refresh token from each response and discard the old one.

With the exit1-mcp npm package

The exit1-mcp package does all of this for you.

Add it to Claude Code with no API key:

claude mcp add exit1 -- npx -y exit1-mcp

Started without EXIT1_API_KEY, the server runs in setup mode and lists a create_account tool. It takes email and an optional agent_name. The agent calls it, the server signs the user up and connects on the spot, and every other tool then works.

From a terminal instead:

npx exit1-mcp signup --email you@example.com

Tokens are saved to ~/.exit1/credentials.json and refreshed automatically. If EXIT1_API_KEY is set, it always wins.

Errors

Every error is JSON with error and error_description.

For an existing account, connect the hosted MCP server with OAuth. The user signs in through their browser:

claude mcp add --transport http exit1 https://app.exit1.dev/mcp/v1 && claude mcp login exit1

The rules

These rules are the safety model. Agents should tell their user about them.

• The account is unclaimed until the user opens the claim email from exit1 and clicks through to app.exit1.dev/claim .
• On the claim page the user chooses to keep or revoke the agent’s access. They can turn on email alerts to their address, and they accept the Terms.
• Until claimed, no alerts are delivered. That covers email, webhooks, Slack and every other channel. Checks are created and run right away, and history builds. Test alerts return 403 with account_unclaimed.
• While unclaimed, GET /account returns "claimed": false plus a claimHint.
• The agent’s token stops working if the user signs in without claiming, if they revoke the agent on the claim page, or when an unclaimed account is paused. Revoking also removes the webhooks and recipients the agent added.
• Unclaimed accounts are paused after 7 days and deleted after 30.
• Accounts start on the Free plan.

Why alerts wait: exit1 can’t let an agent point alerts at an address nobody has proven they own.

Recommended flow for agents

1. If you have no exit1 credentials, ask the user for their email (or use the one they gave). Then POST /oauth/agent-signup.
2. Store the tokens. Use the access token as a Bearer token for the REST API or the hosted MCP server. Or use exit1-mcp, which does all of this through create_account.
3. Set up monitoring: call get_account (or GET /account), read the repo, create checks, and set email alerts to the user’s address with PATCH /alerts/email.
4. Tell the user to open the claim email from exit1 so alerts start. Tell them they choose whether you keep access.

Questions? Email connect@exit1.dev.