Loading ToolBench
Four free MCP tools to check whether AI engines can find, read and recognise a website. No account, no API key, no LLM calls.
Four well-named, read-only tools with strong, domain-specific descriptions (150 - 250 chars each). All tools have identical input schema (url: string) with clear parameter description. Descriptions explicitly state what each tool measures AND what it cannot detect, a pattern rarely seen in community servers. Error handling is thoughtful: the callApi() function converts API errors into actionable human-readable messages rather than stack traces. However, output schemas are not formally documented, responses are JSON text blobs without declared field structure. No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. No pagination or result limits documented. Tool names follow verb_noun pattern well (crawler_watch, entity_check, offpage_gaps, shopping).
Checks whether AI crawlers are allowed to read a site: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Extended. Reads the site's robots.txt and reports, per crawler, whether it is fully allowed, partially blocked or blocked entirely. Use this when someone asks why AI never mentions their site, or before any other AI-visibility work: a blocked crawler makes everything else pointless. Reflects robots.txt only, so it cannot detect blocking done at the CDN or firewall layer.
Checks whether a brand reads as a clear, consistent entity that AI can recognise: Organization schema, a consistent name, a description, a logo and linked profiles (sameAs). AI engines struggle to name a brand they cannot identify as a distinct thing. Use this when a brand is confused with another, or described vaguely, in AI answers. Measures the signals published on the site itself, not the brand's reputation across the web.
Finds where a brand is missing OUTSIDE its own website: review platforms, directories, communities and 'best of' roundups relevant to its category. Off-page mentions on third-party pages are the strongest known predictor of AI citations, because engines weight what others say about a brand far above what the brand says about itself. Use this when a site is technically perfect yet still never cited. Returns the gaps and priorities; it does not publish anything or contact anyone.
Checks whether AI shopping agents such as ChatGPT Shopping can find, understand and recommend a store's products: product schema completeness, price and availability signals, and the attributes those agents read. Use this for e-commerce sites, especially Shopify, when products never surface in AI recommendations. Inspects published structured data, so it cannot see a merchant feed submitted privately to a platform.
Output schemas not formally documented. Responses are JSON text blobs; LLMs cannot infer field structure or plan downstream tool calls. No declared return type schema.
No tool annotations despite all tools being read-only. Missing readOnlyHint and idempotentHint annotations that would signal safety to agents.
No pagination or result limits documented. Tools return unbounded JSON responses; large audits could exhaust context windows.
Error handling is good (human-readable messages) but lacks error classification. LLMs cannot distinguish retryable (rate limit) from fatal (offline site) errors without parsing text.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Timeout handling returns a message but does not suggest retry strategy. LLMs may not know whether to retry immediately or wait.