VPNDetection

VPNDetection MCP Server

The official Model Context Protocol server for the VPNDetection API.

It gives an AI agent seven read-only tools for anonymity detection: whether an address belongs to a VPN, a residential, datacenter, mobile or public proxy, a Tor node, a privacy relay or web gateway, a hosting provider or a CDN, plus the database catalog and where your organization's license for each one stands.

Getting Started

We host it at https://mcp.vpndetection.io/mcp, and you sign in to it with your VPNDetection account. It also runs on your own machine from npm.

In Claude

In claude.ai, the desktop app or Cowork, add https://mcp.vpndetection.io/mcp as a custom connector and choose Use Claude's published identity when asked. In Claude Code, install our plugin, which adds the server with skills:

/plugin marketplace add vpndetection-io/claude-plugin
/plugin install vpndetection@vpndetection

Either way you sign in with your VPNDetection account, and Claude never sees your API key. The steps, the skills and how to disconnect: docs.vpndetection.io/integrations/claude.

In ChatGPT and Codex

In ChatGPT on the web, add https://mcp.vpndetection.io/mcp as a custom MCP server under Plugins and choose OAuth. In Codex, install our plugin, which adds the server with skills:

codex plugin marketplace add vpndetection-io/openai-plugin
codex plugin add vpndetection@vpndetection
codex mcp login vpndetection

Either way you sign in with your VPNDetection account, and neither sees your API key. The steps: docs.vpndetection.io/integrations/chatgpt and docs.vpndetection.io/integrations/codex. On the OpenAI API, the Responses API's mcp tool takes the server as its server_url and your key as its authorization.

In any other MCP client

Point it at https://mcp.vpndetection.io/mcp. A client that supports MCP authorization signs in the same way. One that doesn't can send a key instead, as Authorization: Bearer your-key.

On your own machine

No API key needed to start. Without a key, a lookup answers ip and is_vpn, up to 1000 a day per source address. A free key adds Tor and relay detection, where the address is and the AS that announces it, and every paid plan answers every field.

Add this to your MCP client's config:

{
  "mcpServers": {
    "vpndetection": {
      "command": "npx",
      "args": ["-y", "vpndetection-mcp"]
    }
  }
}

Requires Node.js 22 or newer.

A key unlocks the provider name, the classification databases and the proxy families. Put it in the environment:

{
  "mcpServers": {
    "vpndetection": {
      "command": "npx",
      "args": ["-y", "vpndetection-mcp"],
      "env": { "VPNDETECTION_API_KEY": "your-key" }
    }
  }
}

VPNDETECTION_BASE_URL overrides the endpoint if you need to point somewhere else.

Tools

Tool	What it answers
lookup_ip	Classify one address.
lookup_ips	Classify a whole list of addresses in one call, keyed by address. A long list is batched for you.
my_entitlement	What this key is entitled to and what it has spent: plan, field tier, requests so far, allowance, and when it resets.
list_databases	Every database we publish, with its standing for your organization: licensed, expired or unlicensed.
database_metadata	A database's columns, sample rows, row count, build date and file sizes.
database_checksum	The published digests for one database file.
list_downloads	Your organization's recent download attempts, refusals included.

The database tools need a key with the db.download scope, which your organization's Default key has.

Every tool is read-only. There is deliberately no download tool: the databases run to several GB, which is not something an agent should pull into a conversation. Fetch them with the client libraries or the API instead.

There is deliberately no my_ip tool, although every client library has one. Over a hosted transport the address our edge observes belongs to whatever proxied the call - Claude's infrastructure, not the person asking - so the tool would answer confidently and wrongly for the only reading anyone would put on it. my_entitlement has no such problem and is the same answer from any transport, because it describes the credential rather than the connection. A test pins the tool's absence so it cannot be added back by accident.

Usage counts against the anniversary of the subscription, not the calendar month and not the billing period. A null hard_limit means we never stop serving; it is not a limit of zero.

Reading a result

Each lookup comes back with a coverage block beside it:

{
  "result": { "ip": "45.83.91.1", "is_vpn": true },
  "coverage": {
    "included": ["ip", "is_vpn"],
    "not_included": ["is_hosting", "is_tor", "hosting", "tor", "..."],
    "note": "The fields in not_included were not returned, because this API key's plan does not include them. ..."
  }
}

This matters more here than in a normal client library. A field missing from a result means your plan doesn't include it, never "we checked and found nothing" - and a model reading the result on its own will otherwise treat the absence as a negative answer. coverage states the difference explicitly so it can't.

Other Libraries

There are official VPNDetection client libraries available for many languages including PHP, Python, Go, Java, Ruby, and many popular frameworks such as Django, Rails, and Laravel. See our GitHub at https://github.com/vpndetection-io for more.

About VPNDetection

VPN Detection API: Accurate anonymity detection identifying VPNs, residential proxies, hosting servers, Tor nodes, CDNs, relays and more.

License

This project is licensed under the MIT License.