Skip to content

Vantage MCP server

dev.vantagemcp/vantagevantagemcp.devstreamable-httpv1.14.0Scanned

Checks whether ChatGPT, Gemini and Perplexity cite your site, and who they cite instead.

11 tools: 1 change things and 10 only read.

Tools: 0 delete, 0 irreversible, 1 write, 10 read.
  • Reads private data · 0
  • Reads text others write · 11
  • Can send data out · 1
  • 5 findings

Tools listed without sign-in. Remote posture

With Anlyon

1 tool here changes things. Govern it.

  • Anlyon lists the tools and adopts none.
  • Each tool you adopt gets approvals and receipts.
  • Your agent never holds the key.
Govern this server →

A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.

What the Vantage MCP server can do

The Vantage MCP server is published by vantagemcp.dev, a domain the MCP Registry verified. It has 11 tools. It can create or change data (1 tool: check_page_readiness), but cannot delete data. 10 tools only read.

11 tools bring in text other people can write (get_supported_options, get_usage, find_citation_leaders and 8 more); 1 tool can send data out of your system (check_page_readiness).

The fixed checks matched 5 findings (long-description). Each one comes with the text it matched.

Its remote endpoint lists its tools without sign-in.

Last scanned Nov 14, 2026, version 1.14.0.

Tools

Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.

11 tools

Changes things 1Claude Code: ask
  • check_page_readinessCheck whether one page has what AI answers look for before they cite it, for the question it should answer, and say exactly what to change, before or after a page is edited.Reads text others writeCan send data outNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
    Description, as the server wrote it
    Check whether one page has what AI answers look for before they cite it, for the question it should answer, and say exactly what to change, before or after a page is edited. It reads the HTML a crawler gets, so it also catches content that only appears with JavaScript. Checks: the <title> asks the question and fits (60 characters); one <h1> matching it; the first paragraph after the heading is the answer itself (12 to 45 words, a sentence, with a number or date); a visible "Updated" date that matches the JSON-LD dateModified and is under 90 days old; at least 300 words without JavaScript; valid structured data, and FAQ markup only for questions shown on the page; canonical and noindex; robots.txt letting OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot reach the page. Costs 0 quota units: it fetches the page and its robots.txt itself and never calls the data provider. Read-only, safe to retry. Passing makes a page easier to cite, it does not cause a citation: whether the question is one your site can own matters more, which only analyze_citation_coverage measures. Returns: {"url" (after redirects), "question", "passed", "failed", "fixes" (what to change, one line per failed check, in page order), "checks" (each {"rule", "ok", "observed", "fix"}), "basis" (one sentence on where the checks come from)}, or {"url", "error"} when the page cannot be fetched (not public, not HTML, an HTTP error). Args: url: full address of the page, e.g. "https://example.com/pricing". question: what the page should answer, in the words people search with, e.g. "how much does example cost".

    Annotations the server declares

    • title: Check a page is ready to be cited
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
Only reads 10Show 10 read-only tools
  • analyze_citation_coverageCheck which of several prompts/keywords actually cite a specific domain, and which ones don't.Reads text others write1 finding

    Why it is in this group

    • The name "analyze_citation_coverage" says "analyze": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The description is 3,784 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...ucture. samples: independent answers to read per keyword, 1 to 5. Default 1."

      What this check looks for
    Description, as the server wrote it
    Check which of several prompts/keywords actually cite a specific domain, and which ones don't. It answers "out of everything we care about, where do we already show up, and where are we invisible"; its not_cited list is the set of keywords worth a closer look. Read-only for the caller, safe to retry. Costs 1 quota unit per keyword per sample (1 sample by default; free tier is 30 units/month shared across every metered tool, so up to 30 keyword checks that period if nothing else is used). A per-keyword provider error doesn't fail the whole call - that keyword's entry just carries an "error" field instead. On the hosted endpoint each keyword's result is remembered for 180 days against your API key, so the next check of the same domain and keyword reports what changed (see "previous" and "change"); get_check_history reads that record back for free. Returns: {"domain", "engine", "samples", "keywords_checked" (int, excludes any that errored), "keywords_cited" (int), "coverage_pct" (float, 0-100), "not_cited" (list of the keyword strings where domain was not cited - the actionable list), "newly_cited" and "no_longer_cited" (keywords whose status flipped since your last check of them), "results" (one entry per keyword, in the order given: {"keyword", "cited" (bool: cited in at least half of the samples), "cited_runs" (how many sampled answers cited it), "samples_ok" (how many answers came back), "rank" (int|null, best 1-based position among the answers' sources, null when never cited), "cited_pages" (list of your own page URLs the answers cited, empty when none - saved with the check, so get_check_history can follow one page over time), "num_sources_cited", "source_domains" (who IS cited, for a keyword you are not in), "source_mix" (how much of that is community sites such as Reddit, YouTube, X - where to get discussed to close the gap), "source_frequency" (only with samples above 1: {"domain", "runs"} per domain), "leads_with_list", "opening_word_count", "mentioned" (bool - the answer's text names the domain or brand in at least half of the samples, whether or not it links to it), "model", "checked_at", "previous" ({"checked_at", "cited_runs", "samples", "best_rank"} from your last check of this domain and keyword on the same engine and market with the same number of samples, or null), "change" ("first_check", "up", "down" or "same", comparing citation rates)}, or {"keyword", "error"} for one that failed), "keywords_mentioned" (int), "mentioned_not_cited" (keywords where the answer names you but does not cite you - the model already knows you, it just isn't linking you), "mention_terms" (exactly what was looked for in the answer text)}. "Cited" and "mentioned" are separate claims and are never merged. Answers change from run to run: with samples=1 a single answer decides "cited", so one sample is weak evidence either way. Args: domain: bare domain to check, e.g. "example.com" (no https://, no www). keywords: prompts/topics to check it against, e.g. ["best project management software", "asana alternatives", "free project management tool"]. Max 10. brand: optional brand name to look for in the answer text, e.g. "Notion". Without it, the domain's first label is used ("notion" for notion.so), which can match an ordinary word by accident for a dictionary-word domain, so pass the real brand when known. country: market to read the answers in, e.g. "Italy". Defaults to "United States". language: language code, e.g. "it". Defaults to "en". Write the keywords in that language too. engine: "chat_gpt" (default), "gemini" or "perplexity", as in analyze_citation_structure. samples: independent answers to read per keyword, 1 to 5. Default 1.

    Annotations the server declares

    • title: Check if AI cites your site
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • analyze_citation_gapCompare your own page's structure against the AI-generated answer actually cited for this keyword, and return a fix brief: the differences between the two, each with the change wor…Reads text others write1 finding

    Why it is in this group

    • The name "analyze_citation_gap" says "analyze": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The description is 2,805 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...5. Default 1. Answers change from run to run; 3 gives a steadier brief."

      What this check looks for
    Description, as the server wrote it
    Compare your own page's structure against the AI-generated answer actually cited for this keyword, and return a fix brief: the differences between the two, each with the change worth testing, not just a description of the winner. Use this to answer 'how does my page differ from what AI cites for this, and what could I try'. The brief lists things to test, not proven causes of citation. It never contains the cited answer's text. Read-only: no side effects, safe to retry. Costs 1 quota unit per sample (1 by default; free tier is 30 units/month shared across every metered tool, so up to 30 single-sample calls to this tool alone if nothing else is used that period). A sample that fails is not charged. Returns: {"keyword", "your_url", "already_cited" (bool: your page's site is among the answer's sources, in at least half of them with samples above 1; when true, fix_brief is one line saying so instead of changes), "basis" (one sentence: how many answers, which engine and market, and that the brief lists things to test), "samples_ok" and "cited_in_answers" (only with samples above 1), "winning" (structure of the AI-cited answer, same shape as analyze_citation_structure), "yours" (same structure computed for your_url, including its own "outline" and "has_table", with "num_links_out"/"linked_domains" standing in for source count), "gaps" (list of plain-English differences worth acting on), "possibly_missing" (heads from the winning outline whose key words mostly do not appear on your page; word matching, so check each before adding it), "fix_brief" (list, most important first, each line what was observed then what is worth testing: opening, number, list, sections, missing points, table, then a reminder to write in your own words - or a "no structural change indicated" line when every check already matches - and always last, one off-site option drawn from the winning answer's source_mix: which community sites (Reddit, YouTube, X...) it cites, or which other sites to get mentioned on). With samples above 1, "winning" is the first answer, and a brief line or missing point needs at least half the answers behind it}, or {"error"} if either side couldn't be fetched/parsed. Args: keyword: the topic/query to check, e.g. "best project management tool". your_url: full URL of your own page to compare, e.g. "https://example.com/best-project-management-tools". country: market to read the cited answer in, e.g. "Italy". Defaults to "United States". language: language code, e.g. "it". Defaults to "en". engine: "chat_gpt" (default), "gemini" or "perplexity": whose answer to compare your page against. samples: how many independent answers to read, 1 to 5. Default 1. Answers change from run to run; 3 gives a steadier brief.

    Annotations the server declares

    • title: Compare your page with the cited answer
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • analyze_citation_structureAnalyze the structural shape of the AI-generated answer actually cited for a keyword: does it lead with a list, how long is the opening passage, how many sources does it cite and f…Reads text others write1 finding

    Why it is in this group

    • The name "analyze_citation_structure" says "analyze": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The description is 2,399 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...eb search. samples: how many independent answers to read, 1 to 5. Default 1."

      What this check looks for
    Description, as the server wrote it
    Analyze the structural shape of the AI-generated answer actually cited for a keyword: does it lead with a list, how long is the opening passage, how many sources does it cite and from which domains. Use this to understand what a winning AI-search answer looks like for a topic, e.g. before writing content on it. Read-only: no side effects, safe to retry. Costs 1 quota unit per sample (1 by default; free tier is 30 units/month shared across every metered tool, so up to 30 single-sample calls to this tool alone if nothing else is used that period). Returns: {"keyword", "engine", "model" (the answering model's version, as the provider reports it), "checked_at" (when the answer was fetched, UTC), "leads_with_list" (bool), "opening" (the answer's opening paragraph, as text), "opening_word_count" (int), "opening_has_number" (bool), "detail_preview" (up to 240 characters of what follows the opening, cut at a word), "outline" (list of up to 12 section heads, in order: the answer's headings, or its top-level list items when it has fewer than two headings; heads only, never the text under them), "has_table" (bool), "num_sources_cited" (int), "source_domains" (list of up to 10 domain strings), "source_urls" (the exact cited page URLs, up to 30, so you can see which page of a domain was cited), "source_mix" ({"community_pct" (share of those sources that are community sites such as Reddit, YouTube, X, Quora), "community_domains", "other_domains"}), "country", "language"}. With samples above 1 the shape fields describe the first answer, plus "samples_ok" (answers that came back) and "source_frequency" (list of {"domain", "runs"}: how many of the answers cited each domain, most often first). Answers change from run to run, so a domain cited in every sample is a far stronger signal than one sample. Args: keyword: the topic/query to analyze, e.g. "how to reduce churn". country: market to read the answer in, e.g. "Italy". Defaults to "United States". For perplexity a 2-letter code also works. language: language code, e.g. "it". Defaults to "en". Write the keyword in that language too. engine: "chat_gpt" (default), "gemini" or "perplexity". chat_gpt and gemini are the answers a person sees in those apps; perplexity is Perplexity's sonar API with web search. samples: how many independent answers to read, 1 to 5. Default 1.

    Annotations the server declares

    • title: Analyze a winning AI answer
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • analyze_citation_structure_batchAnalyze the structural shape of the winning AI answer across several related keywords/topics in one call: does each lead with a list, how long is the opening, how many sources it c…Reads text others writeNo findings

    Why it is in this group

    • The name "analyze_citation_structure_batch" says "analyze": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Analyze the structural shape of the winning AI answer across several related keywords/topics in one call: does each lead with a list, how long is the opening, how many sources it cites. It covers a topic cluster in one call, for content planning across related pieces. Read-only: no side effects, safe to retry. Costs 1 quota unit per keyword in the batch (free tier is 30 units/month shared across all the metered tools, so up to 30 keywords total that period if nothing else is used). A per-keyword provider error doesn't fail the whole batch - that keyword's entry just carries an "error" field instead. Returns: {"results" (list, one {"keyword", ...same shape as analyze_citation_structure, or "error"} per keyword, in the order given), "summary": {"topics_analyzed", "topics_requested", "list_led_count", "avg_sources_cited", "avg_community_pct" (average source_mix.community_pct across the analyzed topics)}}. Args: keywords: topics/queries to analyze, e.g. ["how to reduce churn", "churn rate benchmarks", "reduce customer churn saas"]. Max 10. country: market to read the answers in, e.g. "Italy". Defaults to "United States". language: language code, e.g. "it". Defaults to "en". engine: "chat_gpt" (default), "gemini" or "perplexity", as in analyze_citation_structure. One answer per topic.

    Annotations the server declares

    • title: Analyze winning AI answers in bulk
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • analyze_citation_trendTrack how a domain's AI-citation count has moved month over month, so you can see whether visibility is growing or fading instead of only ever checking a single point in time.Reads text others write1 finding

    Why it is in this group

    • The name "analyze_citation_trend" says "analyze": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The description is 2,658 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...nguage code, e.g. "it". Defaults to "en" (the only option for chat_gpt)."

      What this check looks for
    Description, as the server wrote it
    Track how a domain's AI-citation count has moved month over month, so you can see whether visibility is growing or fading instead of only ever checking a single point in time. Use this to answer 'is our AI visibility improving' or 'did that content push actually move the needle'. Pass compare_domains to put competitors' trends next to it. Read-only: no side effects, safe to retry. Costs 10 quota units per domain (the main domain plus each one in compare_domains; free tier is 30 units/month shared across every metered tool, so up to 3 single-domain calls to this tool alone if nothing else is used that period). A domain the provider fails on is not charged. Returns: {"domain", "platform", "months" (list of {"year", "month", "mentions" (int, 0 for a month with no tracked citations. A zero between two large months can be a gap in the provider's history rather than a real drop, so read isolated zeros with care), "ai_search_volume"}, oldest to newest), "trend": {"direction" ("up"/"down"/"flat"/"no_data"), "earliest_mentions", "latest_mentions", "excluded_current_partial_month" (bool, only present and true when the most recent calendar month was excluded from the trend calculation because it is still in progress and its count is not yet final - it is still returned inside `months`, just not compared)}}. The most recent entry in `months` (or `trend.latest_mentions` when the current month is not excluded) already IS the current count, so there is no need for a separate call just to see it right now. With compare_domains the result also carries "comparisons" (one {"domain", "months", "trend"} per extra domain, same shape and month window as above, in the order given, or {"domain", "error"} for one the provider failed on - that one is refunded and the rest still return). Args: domain: bare domain to check, e.g. "example.com" (no https://, no www). compare_domains: optional list of up to 5 more bare domains to track over the same months, e.g. ["asana.com", "monday.com"]. platform: "chat_gpt" or "google" (Google's AI Overview). Defaults to chat_gpt. Perplexity and Gemini aren't available - the underlying data provider doesn't cover them for this check. months: how many recent months of history to return. Defaults to 6, capped at 13 - DataForSEO's historical data only goes back to 2025-08-01. country: market to check, e.g. "Italy". Defaults to "United States". chat_gpt only has data for the United States; use platform "google" for any other country. language: language code, e.g. "it". Defaults to "en" (the only option for chat_gpt).

    Annotations the server declares

    • title: Track a site's AI citations over time
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • find_citation_leadersFind which domains dominate AI-answer citations for a topic/keyword, and optionally check where one domain, or several competitors at once, show up among them.Reads text others write1 finding

    Why it is in this group

    • The name "find_citation_leaders" says "find": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The description is 2,678 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...o", "asana.com", "monday.com"]. Same exact-match rule as compare_domain."

      What this check looks for
    Description, as the server wrote it
    Find which domains dominate AI-answer citations for a topic/keyword, and optionally check where one domain, or several competitors at once, show up among them. Use this to answer 'who's winning AI search for this topic' or 'is my competitor cited more than me for X'. Comparing several domains costs the same single call. Read-only: no side effects, safe to retry. Costs 10 quota units/call (free tier is 30 units/month shared across every metered tool, so up to 3 calls to this tool alone if nothing else is used that period). Returns: {"keyword", "platform", "top_domains" (list of {"domain", "mentions"}, most-cited domains for this keyword/platform, order as ranked by the provider), "top_domains_limit" (int, the provider's own cap on this list - absence from it is NOT evidence a domain has zero citations, only that it did not rank in the top `top_domains_limit`), "compare_domain_rank" (int|null, only present when compare_domain was passed: the domain's 1-based position in top_domains, or null if it did not rank in the top `top_domains_limit`), "compare_domain_ranks" (only present when compare_domains was passed: {domain: rank or null}, same rank rule as compare_domain_rank), "country", "language", "source_mix" ({"community_pct" (share of these mentions that go to community sites such as Reddit, YouTube, X, Quora), "community_domains", "other_domains"}: a high community_pct means this topic is won by what people say about a brand elsewhere, not by any one site's pages)}. This tool's citation universe is the provider's tracked mention corpus for the keyword, which is a different measurement from analyze_citation_structure's single live answer - the two can legitimately disagree on whether a given domain shows up. Args: keyword: the topic/query to check, e.g. "best project management tool". platform: "chat_gpt" or "google" (Google's AI Overview). Defaults to chat_gpt. Perplexity and Gemini aren't available - the underlying data provider doesn't cover them for this check. compare_domain: optional bare domain to look up in the results (exact match against the registrable domain, e.g. "notion.so" will not match "mynotion.so.example.com"). country: market to check, e.g. "Italy". Defaults to "United States". chat_gpt only has data for the United States; use platform "google" for any other country. language: language code, e.g. "it". Defaults to "en" (the only option for chat_gpt). compare_domains: optional list of up to 10 bare domains to rank side by side, e.g. ["notion.so", "asana.com", "monday.com"]. Same exact-match rule as compare_domain.

    Annotations the server declares

    • title: Find who AI cites for a question
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • find_cited_questionsFind the questions people ask AI answer engines where a domain is already cited as a source, most-asked first.Reads text others writeNo findings

    Why it is in this group

    • The name "find_cited_questions" says "find": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Find the questions people ask AI answer engines where a domain is already cited as a source, most-asked first. Starts from the domain, so nobody has to guess keywords first. Use this to answer 'what does ChatGPT already cite us for' or to pick the keywords to feed analyze_citation_coverage and analyze_citation_gap. Read-only: no side effects, safe to retry. Costs 10 quota units/call (free tier is 30 units/month shared across every metered tool, so up to 3 calls to this tool alone if nothing else is used that period). Returns: {"domain", "platform", "country", "language", "total_questions" (int, every tracked question citing the domain, which can exceed the list), "questions" (up to `limit`, most-asked first: {"question", "ai_search_volume" (monthly asks as the provider estimates them), "your_position" (1-based position of the domain among that answer's sources), "source_domains" (who else that answer cites), "last_seen" (when the provider last recorded this answer, UTC)})}. An empty list means the provider's tracked answers do not cite the domain, not that no answer anywhere does. This reads the provider's tracked answer corpus, the same measurement as find_citation_leaders, not a live answer: re-check a question with analyze_citation_coverage to see today's answer. Args: domain: bare domain, e.g. "example.com" (no https://, no www). Subdomains are included. platform: "chat_gpt" (default) or "google" (Google's AI Overview). limit: how many questions to return, 1 to 20. Default 20. country: market, e.g. "Italy". Defaults to "United States". chat_gpt only has United States data; use platform "google" elsewhere. language: language code, e.g. "it". Defaults to "en", the only option for chat_gpt.

    Annotations the server declares

    • title: Find questions AI cites your site for
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • get_check_historyRead back your own earlier analyze_citation_coverage results for a domain, newest first, to show progress over time or to confirm whether a change (a rewrite, a new mention somewhe…Reads text others writeNo findings

    Why it is in this group

    • The name "get_check_history" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Read back your own earlier analyze_citation_coverage results for a domain, newest first, to show progress over time or to confirm whether a change (a rewrite, a new mention somewhere) moved anything. Use this for 'has our citation status changed since last week' without spending units. Pass `page` to follow ONE page's citations over time instead of the whole domain's. Costs 0 quota units: it only reads Vantage's own record of your checks, never the data provider. Results are kept 180 days per API key and are only visible to that key. Returns: {"domain", "keyword" (or null for every keyword), "page" (the URL asked about, or null), "checks" (newest first: {"keyword", "engine", "country", "language", "samples", "cited_runs", "best_rank", "mentioned_runs", "checked_at", "cited_pages" (your page URLs cited in that check, or null for a check saved before page tracking), and with `page` set, "page_cited" (true/false, or null when that check did not record pages))}, and with `page` set also "times_cited" (checks that cited that page) and "checks_with_page_data". Args: domain: the bare domain the checks were run for, e.g. "example.com". keyword: optional, only this keyword's history. limit: how many rows to return, 1 to 200. Default 50. page: optional full URL of one page on the domain, e.g. "https://example.com/pricing". Matched ignoring "www.", the scheme, a trailing slash and any #fragment.

    Annotations the server declares

    • title: Show your past citation checks
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • get_supported_optionsList which AI engines, platforms and markets each Vantage tool accepts, and the per-call limits, so you can pick valid arguments before spending any quota.Reads text others writeNo findings

    Why it is in this group

    • The name "get_supported_options" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    List which AI engines, platforms and markets each Vantage tool accepts, and the per-call limits, so you can pick valid arguments before spending any quota. Use this to answer 'does Vantage cover Gemini in Italy' or 'what can I pass as engine or platform'. Costs 0 quota units - a fixed list, it never touches the data provider. Returns: {"live_answer": {"tools" (analyze_citation_coverage, analyze_citation_structure, analyze_citation_structure_batch, analyze_citation_gap), "engines" (the values for `engine`), "country" (any country name the provider accepts; for perplexity one of "perplexity_countries" or a 2-letter ISO code), "language" (language code), "max_samples", "max_keywords_per_call"}, "tracked_data": {"tools" (find_citation_leaders, analyze_citation_trend, find_cited_questions), "platforms" (the values for `platform`), "chat_gpt_market" (the only market ChatGPT has tracked data for), "other_markets" (where to get the rest), "max_trend_months", "max_compare_domains" ({"find_citation_leaders", "analyze_citation_trend"})}, "default_market" ({"country", "language"})}. A market the provider does not cover comes back as a refunded error, never a charge.

    Annotations the server declares

    • title: List supported engines and markets
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • get_usageCheck how much of this billing period's quota is left, before spending any of it.Reads text others writeNo findings

    Why it is in this group

    • The name "get_usage" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Check how much of this billing period's quota is left, before spending any of it. Use this to answer 'how many checks do I have left' or to decide whether a batch call will fit before running it. Costs 0 quota units - this never touches the paid data provider, it only reads Vantage's own record of what has been used. Returns: {"tier", "period" (YYYY-MM), "units_used", "units_limit", "units_remaining"}. find_citation_leaders and find_cited_questions cost 10 units/call, analyze_citation_trend 10 per domain looked up; analyze_citation_structure (and its batch form, per keyword), analyze_citation_coverage (per keyword) and analyze_citation_gap cost 1, times `samples` where a tool takes it. get_check_history, check_page_readiness and get_supported_options cost 0. stdio/local-dev mode (no HTTP access token) has no metering at all - this returns tier "unmetered" with no real limit in that case.

    Annotations the server declares

    • title: Check remaining monthly units
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true

This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.

Exfil surface

Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.

No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data.

Reads private data0
Reads files, mail, messages, databases, secrets, calendars or repositories.
Can send data out1
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.

Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.

Remote posture

What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.

Tools listed without sign-in

Looked at Nov 20, 2026, 18:53 UTC, unchanged since Nov 20, 2026. No credentials were sent and no tool was called. What is looked at.

Reachable
yes
HTTP status of an anonymous initialize
200
tools/list served without credentials
yes
Tools that change things, listed without credentials
1
Protocol version
2025-06-18
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
https://vantagemcp.dev/.well-known/oauth-protected-resource/mcp
Authorization servers
https://vantagemcp.dev
Legacy HTTP+SSE transport
no
Dynamic client registration
yes
Client ID metadata documents
no
  • WarnThe whole server

    The endpoint served tools/list without credentials, and the list has 1 tool that changes things (1 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://vantagemcp.dev/mcp: tools/list returned 11 tools to an anonymous client, including check_page_readiness (write)

    What this check looks for

Posture changes

  1. Nov 20, 2026
    • Tools that change things, listed without credentials: 0 to 1
  2. Nov 13, 2026
    • HTTP status of an anonymous initialize: 401 to 200
    • tools/list served without credentials: no to yes
    • Protocol version: none to 2025-06-18
    • WWW-Authenticate on 401: yes to not observed
    • resource_metadata in WWW-Authenticate: https://vantagemcp.dev/.well-known/oauth-protected-resource/mcp to none

Findings

Matches of fixed checks, with the text they matched.

5 info

History

Changes to its tool definitions and to its remote endpoint, newest first.

  1. PostureRemote posture changed. Tools that change things, listed without credentials: 0 -> 1.View diff
    • Tools that change things, listed without credentials: 0 to 1
  2. PostureRemote posture changed. HTTP status of an anonymous initialize: 401 -> 200; tools/list served without credentials: no -> yes; Protocol version: none -> 2025-06-18 (and 2 more).View diff
    • HTTP status of an anonymous initialize: 401 to 200
    • tools/list served without credentials: no to yes
    • Protocol version: none to 2025-06-18
    • WWW-Authenticate on 401: yes to not observed
    • resource_metadata in WWW-Authenticate: https://vantagemcp.dev/.well-known/oauth-protected-resource/mcp to none

Install

Pinned to version 1.14.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http vantage https://vantagemcp.dev/mcp
claude mcp add --transport http vantage https://vantagemcp.dev/mcp

Run in a terminal.

Claude Code permissions

Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.

.claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__vantage__get_supported_options",
      "mcp__vantage__get_usage",
      "mcp__vantage__find_citation_leaders",
      "mcp__vantage__analyze_citation_trend",
      "mcp__vantage__analyze_citation_structure",
      "mcp__vantage__analyze_citation_structure_batch",
      "mcp__vantage__analyze_citation_coverage",
      "mcp__vantage__analyze_citation_gap",
      "mcp__vantage__find_cited_questions",
      "mcp__vantage__get_check_history"
    ],
    "ask": [
      "mcp__vantage__check_page_readiness"
    ],
    "deny": []
  }
}

Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as vantage. Review it before you use it.

Capability tags behind 11 rules
  • mcp__vantage__get_supported_options untrusted-input
  • mcp__vantage__get_usage untrusted-input
  • mcp__vantage__find_citation_leaders untrusted-input
  • mcp__vantage__analyze_citation_trend untrusted-input
  • mcp__vantage__analyze_citation_structure untrusted-input
  • mcp__vantage__analyze_citation_structure_batch untrusted-input
  • mcp__vantage__analyze_citation_coverage untrusted-input
  • mcp__vantage__analyze_citation_gap untrusted-input
  • mcp__vantage__find_cited_questions untrusted-input
  • mcp__vantage__get_check_history untrusted-input
  • mcp__vantage__check_page_readiness untrusted-input, outbound

Pin

Keep the decision: fail CI when the tools change, follow every change, or show a badge.

Pin in CI

Fail a build when the tools you reviewed are not the tools the server serves today.

curl -s "https://radar.anlyon.com/api/v1/servers/dev.vantagemcp/vantage/verify?hash=100dd615fbe0810dd0406e440a39f516d88eeb6ba56d19d5ea3fb795ae56c9e2" | jq -e '.changed == false'

Pinned hash 100dd615fbe0, from Nov 14, 2026, 01:18 UTC.

Watch

Get told when its tools change: a new tool, a new description, a new class.

https://radar.anlyon.com/servers/dev.vantagemcp/vantage/changes.xml

An Atom feed. Works in any feed reader or Slack's RSS app.

Badge

Anlyon Radar badge for dev.vantagemcp/vantage
[![Anlyon Radar](https://radar.anlyon.com/badge/dev.vantagemcp/vantage.svg)](https://radar.anlyon.com/servers/dev.vantagemcp/vantage)

Source: MCP Registry · repository. Synced Nov 14, 2026, 02:17 UTC.

Vantage MCP server: questions and answers

Answered from the scan above, not from the publisher's description.

Is the Vantage MCP server safe?

No scan can promise that. What its definitions show: 11 tools, 1 of which can change, send or delete something, and 5 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.

Can the Vantage MCP server delete data?

No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.

Can the Vantage MCP server do things that cannot be undone?

No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.

What tools does the Vantage MCP server have?

11 tools: 1 write, 10 read. check_page_readiness, get_supported_options, get_usage, find_citation_leaders, analyze_citation_trend, analyze_citation_structure, analyze_citation_structure_batch, analyze_citation_coverage and 3 more.

Does the Vantage MCP server need sign-in or OAuth?

Its remote endpoint listed its tools to an anonymous client, without sign-in.

Does the Vantage MCP server send data to outside services?

1 tool can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): check_page_readiness.

Has the Vantage MCP server changed its tools?

Not since Anlyon Radar started tracking it. Last scanned Nov 14, 2026.