Let your agent onboard itself

Most tools make you sign up, create a key, and paste it into your agent. AnyAPI flips that around: the agent onboards itself. It installs the AnyAPI skills, mints its own free trial key, discovers the catalog, and makes the first call - all from a single prompt. No account, no email, no credentials to hand over. When the trial budget runs out, one human approval keeps it going.

Which page do you want? This page is for an autonomous agent that calls AnyAPI at runtime - it provisions its own trial key. If instead a coding agent is integrating AnyAPI into an app you’ll ship, the app is the caller: create a normal dashboard key on your funded account and see Agent skills. Trial keys carry a small budget and expire in 7 days, so they belong to autonomous agents, not shipped code.

The one thing you do

Prompt your agent. That’s the whole setup step:

You have access to AnyAPI, a pay-per-call marketplace for scraping and data APIs.
Read https://getanyapi.com/agent-onboarding/SKILL.md, onboard yourself, and make one
real API call. Report what you did.

No email, no key to paste. The agent takes it from there: it mints a free trial key, runs a call, and tells you in one sentence what it did.

Already have a funded account and want the agent to spend from it directly? Hand it an API key instead and skip the trial - see Agent skills. The trial is for the zero-setup, agent-first path.

What the agent does on its own

Installs the CLI (if it has a shell)

An agent that can run shell commands installs the AnyAPI CLI first. One command installs the skills, mints the free trial key, and writes local credentials:

npx -y anyapi-cli@latest init

anyapi init installs the AnyAPI agent skills, offers MCP registration, and runs the trial signup if no key is available - saving it to ~/.anyapi/config.json. After init, the agent calls the installed anyapi binary directly.

Mints its own free trial key

With no shell (a hosted chat or no-code runtime), the agent posts to the public signup endpoint directly - no human, no account, no email:

curl -X POST https://api.getanyapi.com/agent/signup \
  -H "Content-Type: application/json" \
  -d '{ "label": "my-agent" }'

The body is optional; label just tags the key. The response returns the key once, plus everything needed to upgrade later:

{
  "secret": "aa_live_…",
  "keyId": "42",
  "capUsd": 0.15,
  "expiresAt": "2026-07-17T00:00:00Z",
  "clientId": "…",
  "upgrade": {
    "authorizationServer": "https://api.getanyapi.com",
    "scope": "run balance:read"
  },
  "notice": "This is a free AnyAPI trial key: about $0.15 of requests, it self-expires in 7 days, and no account was created. When it runs out, your human can connect a wallet to keep it going."
}

The agent stores secret as ANYAPI_API_KEY and relays notice to you verbatim.

Starts working immediately

The trial key works right away. The agent lists APIs, reads schemas, checks its USD balance, and runs a real call over MCP, the CLI, or REST - all priced in USD, pay per request.

What it costs you to start

Nothing, and nothing is owed. A trial key carries a budget of about $0.15 of requests (the capUsd in the response), creates no account, and self-expires in 7 days if it is never upgraded. If the agent proves useless, you do nothing: the key just expires and nothing lingers.

When the trial budget is spent, calls return HTTP 402 with error code trial_cap_reached and an upgrade object in the body. That’s the signal it’s time to approve continued spend (below). The unspent trial also stops working after 7 days.

Upgrade past the trial

Continuing past the trial takes exactly one human approval, through OAuth 2.1. The agent never stops to ask you for a key - you approve a spend limit, and it keeps working on the same session.

With the CLI (recommended)
Raw REST or MCP (no shell)
Dashboard key (secondary)

The agent runs:

anyapi connect

This requires anyapi-cli >= 0.3.0. It starts an OAuth flow with a localhost callback and prints a single consent URL. Hand that one URL to the human. They open it, reach the consent screen at getanyapi.com/dashboard/authorize, sign in or sign up, see a receipt of what the agent accomplished on the trial, set a spend limit, and approve. The agent then holds an OAuth access token and keeps working - there is no key to swap by hand.

If the installed CLI predates anyapi connect, use the dashboard-key path below instead.

An agent that cannot run a shell surfaces the upgrade to you directly. The 402 trial_cap_reached response body carries the upgrade coordinates (authorizationServer and scope); the agent relays the link to you so you can sign in and approve continued spend at getanyapi.com/dashboard/authorize.

A human already signed in at getanyapi.com can create a full, non-expiring key directly at getanyapi.com/dashboard/keys and hand it to the agent via the ANYAPI_API_KEY environment variable. New accounts include a one-time $1 credit. This is the path to use on a CLI too old for anyapi connect.

Why this is safe

The trial budget bounds what an unapproved agent can spend - about $0.15, no more.
No account is created and nothing is owed. An unspent trial key expires on its own in 7 days.
You decide if and when to continue. The OAuth approval is where you set the spend limit, after you’ve seen a receipt of what the agent actually did.

Give your agent the live tools too: connect the MCP server so it can discover and run APIs without writing HTTP by hand, or point it at the CLI to keep tool schemas out of its context window.

Getting Started

Sell on AnyAPI

API Reference

Let your agent onboard itself

The one thing you do

What the agent does on its own

What it costs you to start

Upgrade past the trial

Why this is safe

​The one thing you do

​What the agent does on its own

​What it costs you to start

​Upgrade past the trial

​Why this is safe

The one thing you do

What the agent does on its own

What it costs you to start

Upgrade past the trial

Why this is safe