Skip to content
Primetra

Agent verification

Is that visitor really GPTBot?

Anyone can put GPTBot in a user agent. Paste one, add the address it came from, and see whether the claim holds and what kind of agent it is. Free, no sign-up, answered while you wait.

What you get back

Claimed vs verified
Which agent the visitor claims to be, and whether the claim holds.
What kind of agent
A training crawler, a search crawler, a fetch a person asked for, an agentic browser, an assistant or a coding agent.
How fresh
The date of the vendor list the answer rests on, and the method version.

Paste the full User-Agent header from your logs. A single token such as GPTBot works too.

Optional

IPv4 or IPv6, as your logs show it. Used only for this check and never stored. Without it, a claim can only come back as can't be verified.

Why it's hard

A user agent is a claim, not a credential

Knowing which agents are real is the first step to knowing whether AI agents can find, use and pay for your product.

  • Copying a name is free

    Scrapers borrow GPTBot or ClaudeBot to slip past blocks and look respectable in your logs. The header alone can't tell them apart.

  • Many vendors publish nothing

    Some agents come with nothing to check against. The honest answer then is "can't be verified", and we say so rather than guess.

  • The lists keep moving

    Vendors add addresses, rename agents and launch new ones every few weeks. Each answer says which list it used and how old it is.

Four answers

What each answer means

Verified
The claim checks out against what the vendor publishes.
Spoofed
It claims an agent whose vendor publishes where it runs, and it came from somewhere else.
Can't be verified
It claims an agent, but there is nothing to check it against: the vendor publishes nothing, or no address was given.
No agent claimed
The user agent doesn't name any AI agent we track.

Agent reference

Every agent we track

One page per agent: who runs it, what it does, the tokens to look for in your logs and robots.txt, and whether a visit can be verified.

Training crawlers

12 agents

Collect pages to train AI models. Blocking one keeps your pages out of future training, not out of answers.

User-triggered fetchers

11 agents

Fetch a page at the moment a person asks an assistant about it.

Agentic browsers

26 agents

Browse and act for a person: open pages, click, fill in forms, check out.

Assistants

24 agents

The assistant apps themselves, as they reach your site, your API or your MCP server.

Coding agents

19 agents

Read your docs and call your APIs while they write code for a developer.

Other agents

13 agents

Link previews, ad checks and agents whose vendor does not say what they are for.

For developers

The same lookup, as a free API and MCP tool

No key, no sign-up. Rate-limited per caller. Send the User-Agent and, when you have it, the IP; you get the same answer as this page, as JSON.

HTTP API POST /api/public/agents/verify

curl
curl 'https://primetra.ai/api/public/agents/verify' \  -H 'content-type: application/json' \  -d '{"user_agent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; GPTBot/1.3; +https://openai.com/gptbot)","ip":"203.0.113.7"}'

MCP tool verify_agent

Add https://primetra.ai/api/mcp to any MCP client and call verify_agent. No key needed.

Can agents find and use your product?

Verification tells you who is visiting. The free agent-readiness check tells you whether they can find you and use what they find.

Run the free check