What is this user-agent?

Identify the crawler behind a raw User-Agent header: operator, category, robots.txt token and stance, how it can be verified, and what blocking it costs. An unmatched string is reported as unmatched, with a labelled substring heuristic and never an identification.

Copy this

curl -s 'https://www.pathwren.workers.dev/tools/classify-ua?ua=Mozilla%2F5.0%20AppleWebKit%2F537.36%20(KHTML%2C%20like%20Gecko%3B%20compatible%3B%20GPTBot%2F1.2%3B%20%2Bhttps%3A%2F%2Fopenai.com%2Fgptbot)&s=client-dossiers'

GPTBot, OpenAI, ai-training, robots token GPTBot, verification by published ranges — and the caveat that a user-agent is a claim.

Run it in a browser — it answers JSON with access-control-allow-origin: *, so fetch() works from any page with no proxy.

No setup at all

No account, no key, no token, no handshake, no session.
GET (and HEAD). Anything else answers 405 with an Allow header and points at the MCP endpoint.
Cacheable: max-age=3600, a strong ETag, and a 304 when you send it back.
Asked with no parameters at all it answers 200 with its own specification, so walking /tools/ never costs you a 400.
CC0-1.0. Nothing is logged about you beyond the request line this host publishes about every client, including itself.

Parameters

parameter	required	what it is
`ua` · `user_agent`	yes	The raw User-Agent header value, URL-encoded.

The same tool over MCP

This endpoint holds no copy of the logic: it calls classify_user_agent on the ai-crawler-index MCP server in-process, through that server's own JSON-RPC envelope. One implementation, two doors — an answer that changes here changed there.

curl -s https://www.pathwren.workers.dev/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"classify_user_agent","arguments":{}}}'

What it will not do

A user-agent is a claim, not proof. Pair it with /tools/verify-crawler, which is the address half of the same question.
It fetches nothing. No argument here or on the MCP tool behind it takes a URL, and neither makes any outbound request: every answer comes out of files this host already publishes.

Also here

The other 7 endpoints, one page each.
123 clients that actually crawled this host, 2026-08-31 20:58:11Z to 2026-09-02 04:57:09Z UTC, one page each.
/data/observed-clients.json — that whole set in one request.