Free VPN and proxy detection API
One POST returns a verdict, a score from 0 to 1 and the network behind an IPv4 or IPv6 address: VPN, proxy, Tor exit, privacy relay or datacenter. Free, with no key and no sign-up.
curl iplogs.com{
"ip": "203.0.113.7",
"verdict": "vpn_detected",
"score": 0.85,
"is_vpn": true,
"is_proxy": false,
"is_tor": false,
"is_relay": false,
"is_hosting": true,
"vpn_provider": "Mullvad",
"asn": "AS9009",
"org": "M247 Europe SRL",
"city": "Phoenix",
"country": "United States",
"country_code": "US",
"signals": [
"vpn_relay_list",
"vpn_aggregator_cidr",
"dc_ip"
],
"details": "https://iplogs.com/ip/203.0.113.7"
}That checks your own address. curl iplogs.com/8.8.8.8 checks another one, and curl iplogs.com/v1/check/8.8.8.8 returns the full response. Anything that is not a command-line client gets compact JSON unless you add ?pretty. For production code, POST https://iplogs.com/v1/check is below.
- 2,000full checks a day, live probes included
- No capon fast checks after that, from our server lists
- 60requests a minute, bursts of up to 30
- No keyno signup, no card
Limits count per calling address (IPv6 per /64), not per address you look up. Bulk takes 100 addresses a call: three calls at once, then one every 5 seconds.
Quickstart
example: a Tor exitcurl https://iplogs.com/v1/check \
-H 'content-type: application/json' \
-d '{"ip":"185.220.101.34"}'terminalNothing to install
Request body
JSON, every field optionalip- The IPv4 or IPv6 address to check. Leave it out to check whoever sent the request.
mode- "fast" answers from memory in microseconds: every list and network check, no live probes or lookups.
Read only when the visitor's browser calls the API itself, with no ip. Sent with an ip, they are ignored.
timezone- The browser's IANA time zone, e.g. America/New_York, compared with the IP's location.
language- The browser's language, e.g. en-US, compared with the IP's country.
webrtc_ip- An address the browser exposed over WebRTC; a different one is a leak.
user_agent- Read from the request header already. Send it only to override that.
Many addresses: POST https://iplogs.com/v1/bulk-check, up to 100 per call (example below), or paste a list into the bulk checker.
Response, 200
{
"score": 1,
"verdict": "vpn_detected",
"is_vpn": true,
"is_proxy": true,
"reputation": "greylist",
"signals": [
{
"type": "tor_exit",
"weight": 0.95,
"matched": true,
"detail": "…"
},
{
"type": "vpn_aggregator_cidr",
"weight": 0.65,
"matched": true,
"detail": "…"
}
],
"ip_info": {
"ip": "185.220.101.34",
"asn": "AS60729",
"org": "Stiftung Erneuerbare Freiheit",
"country": "Germany",
"country_code": "DE",
"vpn_provider": "Tor"
},
"request_id": "req_6228befb-59f"
}Errors
- 400
- Not an IP address, or a private one (10.x, 192.168.x, loopback, link-local).
- 429
- Over 60 requests a minute. Wait Retry-After seconds, then retry. Header:
Retry-After: 1 - 504
- A full check ran past 30 seconds. Retry, or send "mode": "fast".
Endpoints
https://iplogs.com, no key · OpenAPI 3.1 specGET/curl iplogs.com checks the caller and prints a short summary, pretty-printed for curl, wget, HTTPie, xh and PowerShell (add ?pretty from anywhere else).GET/{ip}The same summary for another address, e.g. curl iplogs.com/8.8.8.8.GET/v1/summaryThe summary as JSON for the caller. /v1/summary/{ip} takes an address.GET/v1/lookup/{ip}IP check: owner name, owner domains, ASN, network range, location and list flags from stored data, in about a millisecond, no probing, no daily quota, 50 requests a second. /v1/lookup checks the caller.GET/v1/checkThe full response for the caller, no body needed. /v1/check/{ip} takes an address.POST/v1/checkOne address in full: verdict, score, every signal and the network. Same response as GET /v1/check/{ip}.POST/v1/bulk-checkUp to 100 addresses in one call. One compact row each, no live probes.GET/v1/ipThe caller's own address and its family. ipv4.iplogs.com and ipv6.iplogs.com each answer over one family.GET/v1/health{"status": "ok"} while the API is up.
Many addresses at once
Post up to 100 in ips. Each row carries what a full check does except the signals, and an address that can't be checked comes back with error while the rest still run. Rows skip live probes, so they return in seconds.
{"ips": ["185.220.101.34", "146.70.116.130",
"8.8.8.8", "10.0.0.1"]}{
"total": 4,
"duration_ms": 3001,
"results": [
{ "ip": "185.220.101.34", "verdict": "vpn_detected", "score": 1,
"is_vpn": true, "reputation": "greylist", "vpn_provider": "Tor",
"country_code": "DE", "asn": "AS60729",
"org": "Stiftung Erneuerbare Freiheit" },
{ "ip": "146.70.116.130", "verdict": "vpn_detected", "score": 1,
"is_vpn": true, "reputation": "greylist", "vpn_provider": "Mullvad",
"country_code": "AT", "asn": "AS9009", "org": "M247 Europe SRL" },
{ "ip": "8.8.8.8", "verdict": "clean", "score": 0, "is_vpn": false,
"reputation": "allowlist", "country_code": "US", "asn": "AS15169",
"org": "Google LLC" },
{ "ip": "10.0.0.1", "score": 0, "is_vpn": false,
"error": "IP must be a public address" }
]
}Before you ship
Send the visitor's IP from your server
Leave ip out and the API checks whoever made the request. Called from your backend, that is your own server.
Browser fields only count in the browser
timezone, language, webrtc_ip and user_agent are read only when the visitor's browser calls the API itself, with no ip. Sent alongside an ip they are ignored. The visitor can edit a browser-side result, so use it for soft decisions.
Over the daily quota is still a 200
Past 2,000 full checks a day, answers come from the fast path with "mode": "fast". No live probes run, so a server only a VPN handshake gives away can score lower. Read mode if depth matters.
Privacy relays are not VPN users
iCloud Private Relay and Cloudflare WARP score like a VPN, but most people behind them are ordinary customers with the setting switched on. Corporate web gateways such as Zscaler are proxies by design, used by a company's staff. To let them through while blocking VPNs, check is_relay before acting on the verdict.
Search engines come from hosting networks
Googlebot, Bingbot and AI crawlers run in datacenters, so their verdict can read suspicious. Check is_crawler before blocking: it is set only for addresses in the operator's published ranges, so a scraper faking a Googlebot User-Agent does not get it.
A missing vpn_provider is not a clean result
Many detections come from ranges with no operator named. Decide on verdict or is_vpn; treat vpn_provider as a label.
Bulk rows skip live probes
A /v1/bulk-check row can score lower than /v1/check for the same address. Re-check the borderline ones one at a time.
An answer can be 30 minutes old
Full checks are cached per address for 30 minutes. A VPN server that went live a minute ago may not show yet.
Private addresses are refused
10.x, 192.168.x, loopback and link-local addresses get a 400. In a bulk call each one comes back as a row with error set, and the rest still run.
Give it 30 seconds, or ask for fast
A first look at an unseen hosting address takes 5–7 s; the API stops at 30 s with a 504. With a hard budget, such as a login, send "mode": "fast".
Response fields
| Field | Type | Meaning |
|---|---|---|
score | number | 0 to 1. Higher means more evidence of a VPN, proxy, relay or Tor exit. Not a probability. |
verdict | string | clean below 0.30, suspicious below 0.60, vpn_likely below 0.85, then vpn_detected. Tor exits and relays use the same levels; vpn_provider says which. |
is_vpn | boolean | True when the score is 0.60 or more. |
is_proxy | boolean | The same test, counting only proxy-type evidence. |
proxy_type | string | Set when is_proxy is true and the evidence says which kind: residential (exit of a residential proxy network), isp (static ISP proxy), public (open proxy anyone can use) or datacenter (a proxy company's own servers). Omitted for Tor and relays, which have their own flags. |
is_tor | boolean | A current Tor exit. |
is_relay | boolean | An iCloud Private Relay or Cloudflare WARP exit, or a corporate secure web gateway: relays many sites choose to allow. |
is_crawler | boolean | Inside a crawler's or uptime monitor's published address ranges (Googlebot, Bingbot, Applebot, OpenAI, Perplexity, DuckDuckGo, Common Crawl, Anthropic, Ahrefs, Pingdom, UptimeRobot and others). Verified by address, not by User-Agent. |
crawler | object | Present only when is_crawler is true: name (e.g. GPTBot), operator (e.g. OpenAI) and purpose: search, ai_training, ai_search, user_fetch (a person asked an assistant to open the page), seo, monitor, archive, crawler or ai. Lets you allow ChatGPT-User while blocking GPTBot. |
is_satellite | boolean | The network serves users over satellite (Starlink, Viasat, HughesNet, OneWeb, inflight and maritime Wi-Fi). Location can be far from the user, and round-trip checks are skipped there, since satellite links split connections the way proxies do. |
is_hosting | boolean | A datacenter, cloud or CDN network rather than a home or mobile connection. |
reputation | string | allowlist (no abuse, nothing hidden), greylist (no abuse, but hidden or hosted) or blocklist (recent attacks or abuse). Separate from the verdict. |
mode | string | "fast" when the answer came from the in-memory path. Absent on a full check. |
signals | array | Each check run: type, weight, matched, and a short detail. Server lists, Tor exits and our own sightings also carry last_seen (UTC): when our data last confirmed the address. Only matched signals add to the score. |
ip_info | object | ip, asn, org, country, country_code, city, reverse_dns, prefix, and vpn_provider: the operator (Mullvad, Tor, Apple) when one is known. |
request_id | string | Quote it when you report a wrong result. |
Limits and errors
per IP · IPv6 per /64| Limit | Free tier | Past the limit |
|---|---|---|
| Requests | 60 a minute | 429 with Retry-After |
| Full checks | 2,000 a UTC day | Still 200, answered from the fast path with "mode": "fast"; no daily cap there |
| GET routes | Share the /v1/check limits | As above |
| Bulk | 100 IPs per call; 3 calls in a burst, then 1 every 5 s | 429 with Retry-After |
| Bad address | – | 400 with {"error": "invalid IP address"} |
| Time | 30 s per full check | 504 with {"error": "detection timed out, please retry"} |
Every charged response carries X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset. During load spikes a check may also be answered from the fast path, so read mode if depth matters. Need more full checks? Email admin@iplogs.com.
What to do with a score
| Score | Level | Suggested action |
|---|---|---|
| 0–29 | Clean | Allow. |
| 30–59 | Suspicious | Allow; add friction only on sensitive actions. |
| 60–84 | Likely VPN | Challenge: verify an email or phone, or require a login. Privacy relays: prefer a light challenge. |
| 85–100 | VPN detected | Challenge or block, per your policy on VPNs and Tor. |
How the score works
- Seven layers look at an address: IP intelligence, TCP/IP fingerprint, TLS fingerprint, round-trip timing, active probing, browser signals, and port and network sanity.
- Strong evidence: a VPN server, a Tor exit, a VPN or proxy range, an open proxy, a Private Relay or WARP exit, a live VPN handshake.
- Weaker evidence: a datacenter or cloud network, unusual routing, a browser time zone or language that doesn't fit the IP.
- The score adds the weights of every matched signal, up to 1. Statements about the network alone (hosting, a VPN-friendly hoster, abuse lists, routing) add at most 0.55 together, so they never reach vpn_likely by themselves; a VPN handshake answered on a home line counts for at most 0.55, since routers run VPN servers for their owners. Two agreeing tunnel measurements (MTU, TTL, round trips) add 0.15, and a live handshake on a datacenter address adds 0.10.
- Measured on 2,831 known addresses (3 November 2026): all 1,722 commercial VPN servers detected, 0.3% false positives on 951 homes, mobile lines, universities, offices, clouds and CDNs. Details on the accuracy page.
- Weak spot: a residential proxy exit is close to undetectable from the IP alone.
Signal types
tor_exit | 0.95 | A current Tor exit relay. |
vpn_relay_list | 0.85 | A server a VPN provider publishes. |
vpn_server_seen | 0.70, then 0.45 | A hosting address answered a VPN or proxy handshake in one of our full checks within 30 days; lets fast and bulk checks, which never probe, carry it. Strong for a week, weaker after. |
vpn_relay_recent | 0.25–0.70 | Was on a provider's published list in the last 30 days but not in the latest copy. Strong on hosting networks, weak and short-lived on home lines. |
known_vpn_ip / known_vpn_cidr | 0.85 / 0.80 | On a curated list of VPN addresses or ranges. |
vpn_aggregator_cidr | 0.65 | Inside a range shared by VPN and proxy services. |
vpn_dense_block | 0.40 | Not itself on a server list, but its /24 holds 40 or more addresses that are (the detail names the main provider): almost always another server of the same fleet. Suspicious alone; with hosting evidence, vpn_likely. |
cloudflare_warp / apple_private_relay | 0.85 / 0.50 | A WARP or iCloud Private Relay exit. |
known_crawler | 0 | A search or AI crawler's published address (the detail names the bot). Weight 0; sets is_crawler. Crawlers run on hosting networks, so the verdict can still read suspicious. |
hosting_stack_mismatch | 0.60 | Self-check only: on a datacenter network, the TCP connection was opened by a different operating system than the browser's, so a proxy server re-opened it. Sets is_proxy. |
hosting_tunnel_mss | 0.50 | Self-check only: on a datacenter network, the browser's own device opened the connection with a tunnel's packet size (MSS 1400 or less), so it is inside a VPN tunnel that exits there. |
dns_leak | 0.50 | Self-check only: on a datacenter connection, the browser's DNS lookup came from a consumer ISP in another country, so the visitor's own network shows through a VPN or proxy that does not carry DNS. Public resolvers never count. |
dual_stack_leak | 0.50 | Self-check only: on a datacenter connection, the visitor's other address family (IPv6 for an IPv4 check, or the reverse) arrives from a consumer ISP in another country, so one family is relayed and the other leaks around it. Hosting networks never count. |
satellite_network | 0 | The network is a satellite internet operator's. Weight 0; sets is_satellite. |
secure_web_gateway | 0 | Egress of a corporate secure web gateway (Zscaler), from the operator's published ranges. Weight 0; proxy observations on the address are kept but weighted 0. |
public_proxy_list | 0.75 (0.40) | An open proxy seen recently. |
residential_proxy_exit | 0.45–0.75 | Seen as the exit of a residential proxy network. |
leased_isp_block | 0.50–0.75 | Inside an address block an IP lessor holds and a home ISP announces, the way ISP (static residential) proxies are set up. Suspicious on its own. |
proxy_operator_network | 0.75 | The network is registered to a proxy company, so the address is one of its datacenter or ISP proxies (the detail names the company). Sets is_proxy. |
active_probe_* | up to 0.70 | Answered a VPN or proxy handshake: OpenVPN, WireGuard, IKEv2, IKEv1/L2TP, SoftEther, SOCKS5, HTTP CONNECT or REALITY. The weight scales with how certain the handshake is. |
vpn_asn / vpn_org_keyword | 0.45 / 0.60 | The network belongs to, or is named like, a VPN provider. |
threat_intel_listed | 0.40 | Reported for attacks or abuse recently. |
dc_ip / asn_cloud_provider | 0.10 / 0.25 | A datacenter, hosting or cloud network. |
tz_mismatch / lang_mismatch | 0.10 / 0.15 | The browser's time zone or language doesn't fit the IP. Browser calls only. |
The middle column is what each signal adds to the score when it matches; every response shows the same weights. Other types can appear. Treat an unknown type as information; the score already counts it.
Integration guides
Support and feedback
Wrong verdict, a bug, or a gap in these docs? Email admin@iplogs.com with the IP and the request_id from the response.
For AI assistants and agents
This page as Markdown: docs.md. Site summaries for language models: llms.txt and llms-full.txt.
Ask an assistant to wire it in:Open in ChatGPTOpen in Claude