Paste your access log. Get back what is crawling you, who runs it, what it costs to block, and which lines are lying about who they are. Streamable HTTP, stateless, no key, no signup. Every tool is read-only.
# three lines of a log, one call, no account
curl -s https://www.pathwren.workers.dev/mcp/triage \
-H 'content-type: application/json' -H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"triage_log",
"arguments":{"log":"66.249.66.1 Googlebot/2.1\n203.0.113.9 Googlebot/2.1\n412 GPTBot/1.2"}}}' \
| jq -r '.result.structuredContent.rows[] | "\(.identified)\t\(.operator)\t\(.verdict)"'
Googlebot Google verified
Googlebot Google impersonation
GPTBot OpenAI unverifiable
With a real file, the shape you actually want:
curl -s https://www.pathwren.workers.dev/mcp/triage \
-H 'content-type: application/json' -H 'accept: application/json, text/event-stream' \
-d "$(jq -Rs '{jsonrpc:"2.0",id:1,method:"tools/call",
params:{name:"summarize_by_operator",arguments:{log:.}}}' < access.log)" \
| jq -r '.result.structuredContent.by_operator[] | "\(.requests)\t\(.operator)\t\(.share_pct)%"'
claude mcp add --transport http crawler-log-triage https://www.pathwren.workers.dev/mcp/triage
Clients that take a JSON config (Claude Desktop, Cursor, VS Code, Windsurf):
{
"mcpServers": {
"crawler-log-triage": { "type": "streamable-http", "url": "https://www.pathwren.workers.dev/mcp/triage" }
}
}
| Tool | What it answers |
|---|---|
triage_log | The whole job. Per line: which crawler, whose, what for, what blocking costs — plus a rollup by operator and category and every line whose address contradicts its claim. |
find_impersonators | Only the lines claiming a crawler whose operator publishes ranges, from an address in none of them. Checked against 1887 IPv4 and 1056 IPv6 prefixes from 12 operator sources. |
summarize_by_operator | Aggregate only: who crawled you, how many requests, what share, which category, what blocking each would cost. Eats a uniq -c table straight from a shell pipeline. |
robots_from_log | A robots.txt naming only the crawlers your log actually contains — not a generic list of everything that exists — with a warning for any that do not documentably obey it. |
waf_ruleset_from_log | nginx, Caddy, Cloudflare, HAProxy or Apache rules for the same set. The enforcement half, for the crawlers that ignore robots.txt. |
Formats are detected per line, so a messy paste is fine and a mixed file is fine:
| Shape | Example |
|---|---|
| combined / common | 66.249.66.1 - - [01/Sep/2026:00:14:02 +0000] "GET / HTTP/1.1" 200 512 "-" "…Googlebot/2.1…" |
| JSON lines | {"remote_addr":"66.249.66.1","http_user_agent":"…Googlebot/2.1…"} |
uniq -c table | 412 Mozilla/5.0 … GPTBot/1.2 … — the count is respected, so 412 lines cost you one |
| ip + user-agent | 66.249.66.1 Googlebot/2.1 |
| bare user-agent | Mozilla/5.0 (compatible; ClaudeBot/1.0; …) |
The usual way to produce the third one:
awk -F'"' '{print $6}' access.log | sort | uniq -c | sort -rn | head -50
ai-crawler-index answers questions about one thing — this user-agent, this IP, this crawler, this stance. It is a reference. This server does not answer questions about crawlers at all: it takes your evidence, a log you already have, and gives back a decision about it. One call in, per-line verdicts plus a rollup plus a config out. Different input, different unit of work, no tool in common — same index underneath, so the two agree by construction. Run both.
It has no side effects and no write path. It has no tool that fetches a URL, and no tool that takes one — it works on text you paste, and nothing else. That is deliberate: a log-analysis tool that will fetch an arbitrary URL for whoever is talking to it is an SSRF proxy with a friendly description, so this one cannot be asked.
Operators who verify by reverse DNS (9 of them, including Bing, Apple and
DuckDuckGo) cannot be checked without a lookup, so instead of doing one it hands you the exact
dig command and says the result is unknown until you run it. A user-agent is a
claim, not a proof; every answer says which of the two you are holding.
Protocol versions 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05, negotiated per call.
server/discover answers for clients on 2026-07-28, initialize for everyone else.
Same index as /data/agents.json, refreshed every six hours.
Listed in the official
MCP Registry as dev.workers.pathwren.www/crawler-log-triage.