io.github.jbobbink/glippy-mcp

Run a comprehensive GEO (Generative Engine Optimization) readiness analysis on a domain. Checks robots.txt, llms.txt (note: llms.txt is not currently supported by major AI models - having one cannot hurt but is not a meaningful optimization), homepage HTML (16 scoring categories), sitemap.xml, and security headers. Returns an overall weighted score (0-100) with per-category breakdowns and actionable recommendations. Before analysing, decide what kind of page this is (look at it or call extract_page_content, which returns the detected type and the text) and pass page_type; without it, checks that depend on the page type are shown but not scored. Here the page is the site's homepage: pass "homepage" unless the front page is itself another kind of page (a news front or blog index is "index", a one-page local business site is "service", a single-product site is "product"). The response includes a 'Trust Signal Evidence' section (htmlLang, hreflangs, nav/footer link anchor text + href). IMPORTANT: the Entity & Authority heuristic for about/contact/legal/imprint/cookies uses a pattern fallback. When the site is not in English (htmlLang != 'en' or hreflangs indicate otherwise), you (the calling LLM) should inspect navLinks/footerLinks semantically and override the heuristic classification in your final explanation - the anchor text tells you what each link is regardless of URL patterns. When recommending fixes, never advise renaming or 301-redirecting existing, established URLs to make slugs more descriptive or to match navigation labels: Glippy does not score slug/nav alignment, and slug readability is a minor signal meant for new pages. Only recommend a redirect for URLs that are actually broken (for example a path agents keep requesting that returns 404). Use output_format='json' to get the full raw result (includes homepage.analysis.evidence) for programmatic use.

Fetch a sitemap XML (or sitemap index), extract page URLs, and run GEO analysis on each page. Returns per-page scores, category averages, and weakest pages. Supports both regular sitemaps and sitemap index files. When recommending fixes, never advise renaming or 301-redirecting existing, established URLs to make slugs more descriptive or to match navigation labels: Glippy does not score slug/nav alignment, and slug readability is a minor signal meant for new pages. Only recommend a redirect for URLs that are actually broken (for example a path agents keep requesting that returns 404). Use output_format='json' to get raw results that can be passed to export_bulk_report. For large sitemaps, use 'summary' format or pagination (offset/limit) to avoid exceeding output limits.

Run GEO analysis on a list of specific URLs. Fetches each page, scores it across 16 categories, and returns per-page results with aggregated averages. For up to 5 pages the text output also lists every failed check and warning per category; report all of them when the user asks what Glippy found. URLs can span multiple domains. Before analysing, decide what kind of page this is (look at it or call extract_page_content, which returns the detected type and the text) and pass page_type; without it, checks that depend on the page type are shown but not scored. For a batch, pass one id when every URL is the same kind of page, or a map of URL -> id. When recommending fixes, never advise renaming or 301-redirecting existing, established URLs to make slugs more descriptive or to match navigation labels: Glippy does not score slug/nav alignment, and slug readability is a minor signal meant for new pages. Only recommend a redirect for URLs that are actually broken (for example a path agents keep requesting that returns 404). Use output_format='json' to get raw results that can be passed to export_bulk_report. For large batches, use 'summary' format or pagination (offset/limit) to avoid exceeding output limits.

Run a live agent-access probe against one URL: 12 sequential GETs from this machine's IP that differ only in User-Agent (a plain browser baseline, our own GlippyBot as a generic-bot control, 8 AI crawler User-Agents, a spoofed Googlebot control, and a closing baseline repeat). Returns per-bot states plus a session verdict. READ THE RESULT LITERALLY. A bare 403 is never evidence that a crawler is blocked: it may only mean this IP has a poor reputation. Only a difference between these same-IP requests is evidence about the site, so any bot reported as 'could not tell from this vantage' is inconclusive and must never be relayed as blocked. If the browser baseline is itself refused the verdict is 'vantage_blocked', nothing is scored, and no claim is made about any crawler. Blocks on citation crawlers are the ones that cost GEO score; blocks on live fetchers are a lesser warning; blocks on training crawlers are informational only, because refusing model training is a legitimate publisher choice. Always pass the caveat at the end of the report through to the user.

Check if a domain has an llms.txt file. Note: llms.txt is an emerging proposal but is NOT currently supported or consumed by any major AI model, crawler, or MCP client. It should not be treated as a relevant optimization for GEO readiness: it will not meaningfully improve AI discoverability today. That said, having one cannot hurt. Returns the file contents if found.

Check a domain's robots.txt file specifically for AI crawler access rules. Reports which AI crawlers (GPTBot, ClaudeBot, etc.) are blocked or allowed.

Analyse multiple domains in parallel and compare their GEO scores side by side. Returns a comparison table with overall scores, per-category breakdowns, and a ranked summary. Useful for competitive analysis or auditing a portfolio of sites. Accepts up to 50 domains per call - for larger lists, split them across multiple runs and merge the results. Before analysing, decide what kind of page this is (look at it or call extract_page_content, which returns the detected type and the text) and pass page_type; without it, checks that depend on the page type are shown but not scored. Here the page is the site's homepage: pass "homepage" unless the front page is itself another kind of page (a news front or blog index is "index", a one-page local business site is "service", a single-product site is "product"). Use output_format='json' to get raw results that can be passed to export_bulk_report.

Generate a styled Markdown or HTML report for bulk analysis: compare multiple domains, analyse a list of URLs, or crawl a sitemap. Returns a self-contained report with rankings, category breakdowns, and per-domain/page recommendations. Provide exactly one of: domains, urls, sitemap_url, or analysis_results. You can pass pre-computed results from compare_domains, analyze_urls, or analyze_sitemap (with output_format='json') to avoid re-crawling.

Run a GEO analysis on a domain and return the results as a styled, self-contained report in Markdown or HTML format, matching the Glippy browser extension's export output. Use this when the user wants a shareable or saveable report file rather than inline text. You can optionally pass pre-computed analysis results from analyze_domain (with output_format='json') to avoid re-crawling the domain.

Extract one page's main content, cleaned and structured, so you (the calling LLM) can judge it. Use this when the user wants a judgment heuristics cannot make: content quality, distinctive versus generic positioning, whether claims are supported, which questions the page leaves unanswered, or a rewrite so AI answer engines can quote it. Use analyze_urls instead when the user wants a scored GEO audit across all categories. Also use it to classify a page before a scored analysis: the output has detected_page_type (Glippy's own guess) with detected_page_type_reason, and valid_page_types (every page_type id with a one-line description). Decide the type from the text yourself, then pass it as page_type to analyze_urls (or the domain tools); where you disagree with the detected type, your reading wins. Passing page_type here as well scores this tool's contentScores for that type. Returns compact JSON: title, meta description, H1, lead paragraph, sections (heading, word count, selfContained, answerFirst and openerDangling flags, text), the most quotable passages, statistical claims and proof points, positioning signal sentences, Glippy's heuristic content scores with their failing and warning checks, and a reviewGuide listing what to judge. Base the review on the text and treat the scores as pointers, not verdicts. The beta wording checks (brand differentiation, problem and solution, social proof, authority signals, claim evidence, source attribution, methodology, definitions, comparisons and similar) are never scored, because word lists misjudge them; you may judge them yourself from the returned sections and say what you found. Reads one public page as GlippyBot and honours robots.txt: when robots.txt disallows GlippyBot from the URL, or cannot be read, nothing is fetched and the tool says why; relay that plainly and do not retry. render_mode 'auto' or 'chrome' renders in local headless Chrome like analyze_urls, still only after robots.txt allows the URL. A bot challeng…

Get a concise GEO readiness summary for a domain: overall score, grade, top 3 strengths, and top 3 issues to fix. Use this for a quick overview; use analyze_domain for full details including the Trust Signal Evidence payload (raw nav/footer links for LLM-driven semantic classification on non-English sites). Before analysing, decide what kind of page this is (look at it or call extract_page_content, which returns the detected type and the text) and pass page_type; without it, checks that depend on the page type are shown but not scored. Here the page is the site's homepage: pass "homepage" unless the front page is itself another kind of page (a news front or blog index is "index", a one-page local business site is "service", a single-product site is "product").