The observation layer for AI agents. Outrings provides deterministic web intelligence through MCP, transforming websites into structured, evidence-backed signals that AI systems can inspect and trust.
Scored across 8 tools
Most tools have clearly distinct roles, and the descriptions effectively establish a depth gradient from get_score to audit_website to check_category. The main overlap is between check_ai_visibility and check_category with category 'ai', which both concern AI crawler access/readability, though the former is more detailed and technical.
Names use snake_case and are mostly verb_noun patterns such as audit_website, check_category, compare_websites, get_score, and get_site_inventory. The only minor deviation is what_changed, which reads more like a query phrase than a verb_noun action.
Eight tools is well-scoped for a website intelligence/audit server, with each tool covering a distinct depth or workflow. There is no obvious bloat, and the set avoids being too thin for the domain.
The surface covers broad audits, targeted category checks, AI visibility, score summaries, competitive comparison, site inventory, badge embedding, and change tracking. This is a comprehensive read-oriented workflow plus the necessary snapshot write for historical comparison, with no major dead ends for website auditing and monitoring.
Run the full 216-check audit of a website and return the overall score, every category score, and the ranked list of what to fix first. Use this when the user asks a broad question about a site: "how is my website", "what is wrong with it", "is this site any good", or when you need grounding before advising on changes. Takes roughly 2-8 seconds. If the user only cares about one area, check_category is faster and returns less.
Inputs: url (required), detail. Returns: url, grade, score, counts, +3 more
Measures technical crawler access and the page signals that affect whether an AI system can retrieve and parse a site. Reports, per crawler, whether GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Google-Extended, Googlebot, Bingbot, Applebot-Extended and CCBot are allowed or blocked, naming the exact robots.txt rule responsible. Also checks whether content is served in HTML rather than requiring JavaScript, whether structured data is present and valid, and whether llms.txt exists. It also requests the page as Googlebot, Bingbot, OAI-SearchBot, PerplexityBot, GPTBot and ClaudeBot and compares each response with what a browser receives, reporting the robots.txt policy and what the server actually returned as separate columns. Crawler identity is declared in the User-Agent and not proven, so a refusal may not apply to the real crawler. It does NOT determine whether an assistant will rank, select, mention or cite the page: access is a precondition for citation, not a cause of it, and the crawlers serve different purposes: for OpenAI, OAI-SearchBot governs eligibility for ChatGPT search while GPTBot relates to model training. Use this to rule out or confirm a technical blocker, and say plainly that ranking and selection are outside what any measurement here can establish.
Inputs: url (required). Returns: url, name, score, counts, +4 more
Inspect a single area and return its verdict, the checks that failed, why each matters and how to fix it. Categories: seo, ai (whether AI crawlers and answer engines can read and cite the site), security (headers, exposure, secrets in source), privacy, tls (certificate and protocol), dns (records, SPF, DMARC, CAA), a11y (accessibility), content, slop (AI-generated-filler signals), design, infra, tech, credibility, correlation (does the page deliver what its title promises), faq, discoverability, local, contact. Use this rather than audit_website when the question is narrow: "is my site secure", "why is my SEO bad", "can ChatGPT see my site".
Inputs: url (required), category (required), failing_only. Returns: url, name, score, counts, +4 more
Score two to four sites and return them ranked, with the category gaps between them. Use it for "how do I compare to my competitors" or "which of these is better built". Each site is a real audit, so this takes several seconds per URL.
Inputs: urls (required). Returns: ranked, results
Produce the HTML (and Markdown) for a badge the site owner can paste into their footer or README. It renders a live SVG showing a real measured score that refreshes every six hours and links through to the full audit, so a visitor can verify it. Use this when someone wants to display that their site was checked.
Inputs: url (required), style, theme, metric. Returns: host, html, linkUrl, imageUrl, +1 more
The cheapest useful call: overall score, letter grade and every category score, with no findings or prose. Use it when comparing sites, when the user only wants a number, or as a first step before deciding which category to look into.
Inputs: url (required). Returns: url, grade, score, auditedAt, +1 more
Return a complete inventory of what the Outrings collector observed on a page: not a sample of it: for one kind of thing: internal links, external links, sitemap URLs, email addresses, scripts, images, forms, iframes, stylesheets or API endpoints referenced. The collector does not execute JavaScript, so this is everything present in the served HTML and the responses it fetched, which is not necessarily everything a real browser would end up loading at runtime. Use it for "what third-party scripts am I loading", "which email addresses are exposed on my site", "how many images are missing alt text", "what does my page link out to".
Inputs: url (required), inventory (required), limit. Returns: url, rows, total, returned, +1 more
Re-audit a site and report what moved since it was last checked: which findings are new, which were resolved, and how each category score shifted. Use it after the user says they fixed something, or to verify that a deployment did not regress anything. Returns comparable=false when the site has no earlier snapshot, which is not an error: it means history starts now. Unlike the other tools this one WRITES: it stores a durable snapshot of the result, because a comparison against history requires contributing to it. The baseline is the most recent retained snapshot and snapshots are kept at most once per site per six hours, so baselineAt may not be the deployment immediately before this one: read baselineAt rather than assuming it is the last change. Check comparable before attributing any score movement to the site: when the ruleset differed between the two measurements it is false, and the difference then includes a change in grading rather than a change in the website.
Inputs: url (required). Returns: to, from, host, note, +16 more
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Add one secure layer between your agents and this server.
