AI Crawlability Check

Check if AI crawlers like GPTBot and ClaudeBot can reach, index and quote any web page.

Hosted MCP Server

npx add-mcp 'https://mcp.milan-novotny.com/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

AI Crawlability Check

AI Crawlability Check icon

AI Crawlability Check is a free Model Context Protocol (MCP) server that tells your AI assistant whether AI crawlers can reach a web page, whether the page can be indexed, and whether its content is quotable.

It checks robots.txt rules for 29 AI crawlers, including GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot, and runs 70 measured checks across four tools. It is hosted, needs no account and no API key, and stores nothing.

Connect

Claude (web and desktop). Settings, Connectors, "Add custom connector", then paste the endpoint URL.

Claude Code

claude mcp add --transport http ai-crawlability https://mcp.milan-novotny.com/mcp

Cursor and other clients with a JSON config

{
  "mcpServers": {
    "ai-crawlability": { "url": "https://mcp.milan-novotny.com/mcp" }
  }
}

Then ask in plain language:

  • "Are we blocking GPTBot, ClaudeBot or PerplexityBot on https://example.com/pricing?"
  • "Check whether this page can be indexed by AI search engines, and list the fixes."
  • "Run all the crawlability checks on our five most important URLs and give me one table."

Tools

All five tools are read-only.

ToolInputWhat it returns
check_ai_crawler_accessurlAllowed or blocked for 29 AI crawlers with the deciding robots.txt rule, plus 9 scored checks: robots.txt health, catch-all rules, server refusals, noindex, snippet rules, declared sitemaps, Content-Signal lines and llms.txt.
check_ai_indexabilityurl15 scored checks: HTTP status and redirects, noindex, canonical target, copy present in the raw HTML, JavaScript dependence, XML sitemap coverage, response time, HTML size, structured data, dates and language.
check_ai_citabilityurl22 scored checks on answer-first structure, question headings, lists and tables, statistics with sources, named entities, schema, author and freshness. Tuned for articles and guides.
check_seo_basicsurl, keyword (optional)24 on-page SEO checks: title, meta description, H1, heading order, keyword use, slug, canonical, links, alt text and schema.
get_review_rubrictype: geo or seoA review rubric as text, for the judgment a script cannot make.

Every URL tool returns a 0 to 100 score, the measured value behind each check, the reason the check matters and a fix for each failure.

Example result

check_ai_crawler_access on a page whose robots.txt blocks most AI crawlers (shortened):

{
  "tool": "check_ai_crawler_access",
  "score": 45,
  "verdict": "Weak",
  "summary": "9 of 19 search and answer-time bots and 3 of 10 training bots may fetch this URL according to robots.txt.",
  "top_fixes": [
    { "check": "answer_bots_allowed", "status": "fail", "fix": "Allow these user agents for this path in robots.txt: ..." }
  ],
  "bots": [
    { "bot": "OAI-SearchBot", "owner": "OpenAI", "role": "answer", "allowed": false, "decided_by": "Disallow: / (group OAI-SearchBot)" }
  ]
}

Crawlers checked

Names and purposes were verified against each vendor's own crawler documentation on 19 November 2026. Bytespider has no vendor documentation and is included because it appears in most block lists. Corrections are welcome as issues or pull requests against src/robots.js.

Search and answer-time crawlers

CrawlerOwnerWhat it is forVendor position on robots.txt
OAI-SearchBotOpenAISurfaces sites in ChatGPT searchHonors
ChatGPT-UserOpenAILive fetch when a ChatGPT user or custom GPT asksMay not apply (user-triggered)
Claude-SearchBotAnthropicCrawls to improve search results in ClaudeHonors
Claude-UserAnthropicLive fetch when a Claude user asksHonors
PerplexityBotPerplexityPerplexity search index; not used for model trainingHonors
Perplexity-UserPerplexityLive fetch when a Perplexity user asksGenerally ignores (user-triggered)
GooglebotGoogleGoogle Search, including AI Overviews and AI ModeHonors
BingbotMicrosoftBing index, which also grounds CopilotHonors
ApplebotAppleSearch across Siri, Spotlight and SafariHonors
DuckAssistBotDuckDuckGoReal-time fetch for DuckDuckGo AI answers; no trainingHonors
meta-webindexerMetaMeta AI search results and citationsNot stated
meta-externalfetcherMetaUser-requested link fetches for Meta AIMay bypass (user-triggered)
Amzn-SearchBotAmazonSearch in Amazon products such as Alexa; no trainingHonors
Amzn-UserAmazonLive fetch for Amazon user requests; no trainingMay not follow all rules (user-triggered)
MistralAI-IndexMistralMistral search index; no trainingNot stated
MistralAI-UserMistralLive fetch when a Le Chat user asksHonors
YouBotYou.comYou.com search indexHonors
Google-AgentGoogleGoogle-hosted agents browsing for a userGenerally ignores (user-triggered)
Claude-AgentAnthropicHosted Claude browser acting for a userNot stated

Training crawlers and control tokens

CrawlerOwnerWhat it is forVendor position on robots.txt
GPTBotOpenAIOpenAI model trainingHonors
ClaudeBotAnthropicAnthropic model trainingHonors
Google-ExtendedGoogleControl token for Gemini training and grounding; no effect on SearchHonors
Applebot-ExtendedAppleControl token for Apple AI training on Applebot dataHonors
meta-externalagentMetaTraining Meta AI models and improving productsNot stated
AmazonbotAmazonImproves Amazon products; data may train Amazon AI modelsHonors
MistralAI-TrainingMistralMistral training datasetsHonors
CCBotCommon CrawlOpen web archive, widely reused as AI training dataHonors
AI2BotAllen Institute for AITraining open language modelsNot stated
BytespiderByteDanceByteDance crawler, reported as model trainingNo vendor documentation found

robots.txt matching follows RFC 9309: the most specific user-agent group wins, the longest matching rule wins, and Allow beats Disallow on a tie.

Limits

  • It queries no index. The scores say whether a page can be crawled, indexed and quoted. They do not say whether any AI engine cites it today.
  • Bot-specific firewall rules are invisible. The server reads robots.txt, meta tags and headers. It reports when its own request is refused, and it never sends another company's crawler user agent.
  • No JavaScript. Pages are read as raw HTML on purpose, because that is how most AI crawlers read them.
  • Sitemaps are sampled. At most 5 sitemap files are read per check.
  • Public pages only. Private, local and non-standard addresses are refused.

How it works

The server is one Cloudflare Worker with no dependencies, no storage and no language model. It implements the MCP Streamable HTTP transport statelessly: every POST to /mcp is a self-contained JSON-RPC exchange.

FilePurpose
src/index.jsMCP protocol handling, tool definitions, landing page
src/crawl.jsCrawler access and indexability tools
src/citability.js, src/seo.jsCitability and SEO tools, scoring
src/robots.jsCrawler list and RFC 9309 matching
src/html.jsDependency-free HTML extraction
src/fetchpage.jsFetch guard: public hosts only, 1.5 MB cap, 12 second timeout
src/rubrics.jsThe two review rubrics
test/run.mjsProtocol, unit and tool tests

Run it yourself

Requires Node.js 22 or later.

npm test

Run the URL tools against a live page:

node test/run.mjs https://example.com/ "optional keyword"

Start a local server on port 8787:

npm run dev

To deploy your own copy, remove the routes line from wrangler.toml (it points at this project's domain) and run npm run deploy.

Privacy

The server keeps no state. It has no database, no accounts and no cookies, and it does not log tool inputs. For a URL it requests that page and the same site's robots.txt, llms.txt, llms-full.txt and up to 5 sitemap files, with the user agent milan-novotny-mcp/1.0. Full policy: https://www.milan-novotny.com/tools/ai-crawlability-mcp/privacy/

Author and licence

Built by Milan Novotný as a personal project. Crawler and product names belong to their owners; this project is not affiliated with or endorsed by any of them.

Released under the MIT licence.