AI agents

Give an AI agent FFraud's IP, email and domain checks as tools: ready definitions for Claude and OpenAI, the calls behind them, a prompt and llms.txt.

An agent that signs people up, answers support or reviews orders can check an address before it trusts it. FFraud's checks are free and need no key, so they go into an agent's tool list as they are. The quickest way in is the MCP server; this page is for agents built on a model's API directly.

#Tool definitions for Claude and OpenAI

Three tools, one argument each. Paste the list for your model's API, and run the call beside each tool when the model asks for it.

[
  {
    "name": "ip_check",
    "description": "Check an IP address with FFraud. Returns a fraud score from 0 to 100, the risk band, whether it is a VPN, proxy, Tor exit or datacenter, the network and country it belongs to, threat tags, and the reason in plain words. Call it before trusting a signup, login or payment from an address.",
    "input_schema": {
      "type": "object",
      "properties": {
        "ip": {
          "type": "string",
          "description": "An IPv4 or IPv6 address, like 185.220.101.34"
        }
      },
      "required": [
        "ip"
      ]
    }
  },
  {
    "name": "email_check",
    "description": "Check an email address with FFraud. Returns whether it is a throwaway (disposable) inbox, a role address, a known spammer or a real provider, with the reason. Call it before accepting a signup.",
    "input_schema": {
      "type": "object",
      "properties": {
        "email": {
          "type": "string",
          "description": "An email address, like someone@gmail.com"
        }
      },
      "required": [
        "email"
      ]
    }
  },
  {
    "name": "domain_check",
    "description": "Check an email domain with FFraud. Returns whether it is disposable or phishing, its risk level and the reason.",
    "input_schema": {
      "type": "object",
      "properties": {
        "domain": {
          "type": "string",
          "description": "A domain, like mailinator.com"
        }
      },
      "required": [
        "domain"
      ]
    }
  }
]

#Where an agent should check

Before it creates an account: the email and the IP address the signup came from.
Before it approves an order, a refund or a free trial: the buyer's IP address.
When it reads logs: the addresses behind failed logins and odd requests.
Before it follows or shares a link from a stranger: the domain.

#A system prompt for fraud checks

System prompt

You can check IP addresses, emails and domains with FFraud (tools ip_check, email_check, domain_check).
Before you trust a signup, a login or a payment, check the address it came from and the email it gives.
Treat a fraud_score above 60 as high risk and above 85 as critical. Say why, in FFraud's own reason.
Weigh a VPN or a datacenter with the other signals before you block anyone.

A VPN or a datacenter alone rarely means fraud: many real people use both. The prompt asks the model to weigh them with the other signals and to quote FFraud's own reason.

#Docs written for a model

File	What it holds
/llms.txt	What FFraud is and where every page is, written for a model.
/llms-full.txt	The whole documentation in one file, for a model to read at once.
/openapi.json	Every endpoint, parameter and answer, for tools that read OpenAPI.

Every docs page also carries a Copy for AI button, which copies that page as Markdown for a chat.

Do AI agents need an API key for FFraud?

No. Single checks need no key. A free key raises the limits and adds batch lookups.

Which models can use these tools?

Any model that calls tools. The definitions here are in Claude's and OpenAI's shapes, and the calls behind them are plain HTTP, so other frameworks take them the same way.

Should an agent block every VPN?

No. A VPN raises the score to the middle band, which calls for a second step such as a CAPTCHA. Block the critical band, 86 and up.

PreviousMCP server NextOpen data