Ayzeo

Aggregates the sources behind the AI answers: which domains the AI models' web searches surfaced across ALL the project's prompts in the window (latest run per prompt+model). Returns up to 60 domains ranked by citations, each with class ('own' = the project's own domains, 'pinned' = a tracked competitor, 'other' = third-party such as directories, review sites, press — the quickest listing wins), citations, prompts (distinct prompts it appeared for), answers (distinct answers whose search results included it) and answerShare (answers as a % of answersTotal, the number of latest answers in the window), models, and sourceType where classified, plus a mix summary, totalCitations and answersTotal. Pass `domain` to drill into one domain: its most-cited URLs (title, citations, answers, answerShare, path bucket), the detail's own answersTotal denominator, and the prompts driving them. Pass `view: "pages"` for the page-level list instead: URLs across every domain ranked by citations (default 100 rows, `limit` up to 300; fields url, domain, title, class, sourceType, citations, prompts, answers, answerShare, models, bucket, unresolved for opaque Google redirect links) plus totalUrls (distinct URLs before the cap) and a `buckets` page-type mix computed over the full cached top-300 list, not only the rows returned under `limit`. Pass `url` to drill into one exact page: its citations, answerShare and the prompts whose answers consulted it (takes precedence over `domain`). Windowed (default 30d; presets up to "all", or startDate+endDate). Use for "which sources do AI answers rely on", "where can I get listed", and content-gap analysis against competitor pages.

get_citation_sources

Returns the project's already-generated content pages with their query, template, enhancement state, placement status, SEO HTML, and edit URLs. Use before recommending a new page to check whether one already exists.

get_content_pages

Fetches the latest stored advanced website/technical analysis for the project. Returns scorePercent plus results.checks (per-check booleans across SEO, structured data, AI readiness, robots/llms.txt, security; false = open issue, null = not measured), results.performance (LCP/CLS/TBT), results.scannerDetails (per-check evidence), results.detailedRecommendations, and results.suggestedSnippets (ready-to-paste JSON-LD, llms.txt, and meta-tag suggestions generated for this site). Check createdAt for staleness before deciding whether to trigger run_website_analysis.

get_analysis_results

Returns project name, brand name (the own brand every mention check reads), domain, plan tier, and basic settings.

get_project_info

Returns aggregated project stats. Windowed (default 30d; pass timeRange or startDate+endDate — two calls with shifted windows give period-over-period comparisons): ownBrand summary (mentionRate, urlCitationRate = real citation rate, visibilityShare, avgMentionPosition, sentiment counts), brandRanking (top 20 brands incl. competitors — re-sort by mentionRate client-side for "top competitors by mention rate"; list order is the project's ranking config), and brandRankingByModel (per-model slices, top 10 each — use for per-model breakdowns; visibilityShare there is computed within that model's slice, not the merged list). All-time: totalPrompts, totalCitationResults, totalBrandMentions, byModel raw run/citation counts, latest analysis score. avgMentionPosition = own brand's rank among brands mentioned in AI answers, lower is better, branded prompts excluded.

get_project_stats

Project-wide summary with two halves whose counting differs on purpose. `pages` is ALL-RUN: every stored answer in the window counts (same-run retries once), unlike get_citation_sources, which reduces to the latest run per prompt+model. It lists the most- and least-visited pages (visited layer = URLs the AI actually opened; distinct from the cited/search-results layer), each row with url, domain, class (values "own", "pinned" = a tracked competitor, "other"), visits (total opens), answers (distinct answers that opened THIS url), answerShare (that answers count as % of answersWithVisits, the answers that opened any page at all; Google AI Overview never exposes visited pages, so answersTotal is the wrong denominator), prompts, models and path bucket, plus totalUrls and the least-list floor (minAnswers: opened in at least 2 distinct answers). `brands` is NOT all-run: it reuses the same curated period-average ranking the dashboard brands table shows (per-day share of prompts mentioning the brand from the day's latest runs, averaged over days with results and across models), resorted into most- and least-mentioned; the least list requires mentions for at least minPrompts distinct prompts so one-off names are not the headline. Brand rows carry mentionRate (rounded to one decimal, so a rarely mentioned brand can read 0 while promptsMentioned is positive), avgMentionPosition, promptsMentioned, visibilityShare, sentiment and isProjectDomain. Windows: presets incl. 6m/1y/all (all = full stored history, window starts 2020-01-01), or startDate+endDate; `model` narrows to one AI model. Use for "which pages do AI models visit most/least", "which brands dominate or barely appear", and period-over-period report tables (two calls with shifted windows).

get_project_summary

All-run analytics for ONE tracked prompt in a window (default 30d): every successful stored run, same-run retries counted once, NOT the latest-run-per-model snapshot list_prompts returns. Returns searchQueries (the exact web searches the models ran while answering — the query fan-out — deduplicated per answer), recurringPhrases (2- and 3-word phrases DERIVED from those queries and appearing in at least two answers: analysis output, never searches anyone ran), recurringBrands (brands named in the answers, counted once per answer and grouped like the Brand Industry Ranking, so spellings and sites of one brand are one row; each flagged isOwnBrand; a competitor's context-only mentions are excluded, while the own brand counts every answer with brand detection that names it, as ownBrandMentioned does; branded prompts contribute no position), a coverage-aware summary, per-model run counts, and a paged answer archive. Read the shares carefully: shareOfAnswersWithQueryCapture and shareOfAnswersWithBrandDetection are percentages of the answers where that field was CAPTURED (the sibling answersWithQueryCapture / answersWithBrandDetection is the denominator), not of every run — a query seen in 3 of 3 answers that captured queries is 100% of those, even if 30 runs exist. capturedQueries is null when the surface cannot expose the fan-out and [] when capture worked and the model ran no search; never read null as zero. Answers are paged 20 at a time (pass page); each row's resultId identifies a stored run, and get_prompt_answer returns the full answer text of up to 20 of them (one page) per call. Use it for 'what do the models actually search for this prompt', 'which brands keep showing up in this prompt's answers', and before recommending content for one prompt.

get_prompt_insights

The web searches the AI engines ran before answering, across EVERY tracked prompt of the project in a window (default 30d): every stored run, same-run retries counted once, each query counted once per answer. Returns searchQueries (the exact queries with answers, prompts that triggered them, shareOfAnswersWithQueryCapture, models, first/last seen and the brands named inside the query text, your own brand flagged isOwnBrand), recurringPhrases (2- and 3-word phrases DERIVED from those queries and seen in at least two answers: analysis output, never searches anyone ran), a coverage-aware summary and a per-model breakdown. Read the shares carefully: they are percentages of answersWithQueryCapture (answers whose own searches were stored), not of every run. Perplexity and Google AI Overview do not expose their searches, so their answers are eligible without query capture. This is the project-wide counterpart of get_prompt_insights' searchQueries. Use it for 'what do the models actually search for across my prompts', to pick the queries and phrases new content should target, and with promptIds to compare prompt groups.

get_search_queries

See what your brand is praised and criticized for, using statements extracted from stored AI answers. Defaults to your own brand and the last 30 days. Basis: latest answer per prompt and engine in the window, including branded prompts; a brand domain selects the statements folded onto its Brand Industry Ranking row on the same basis. Themes come from grouped evidence keywords with bilingual labels; ungrouped keywords are listed raw (labelEnglish is null). Returns summary counts, themes per sentiment with total and returned, top member keywords and up to 3 sample statements per theme. Reads stored answers only and spends no runs or quota.

get_sentiment_themes

Returns the full stored answer text of up to 20 past runs, by id: the resultId values of get_prompt_insights' answer archive (20 per page, so one page is one call) or the id values of get_citation_history. Each answer carries its prompt, model and executedAt plus the run fields list_prompts returns for a latest run (brandMentioned, mentionPosition with mentionedBrandCount, sentiment, mentionedBrands and mentionedBrandSentiments, citedUrl, citedUrls, visitedUrls, searchTerms). answer is null when that run stored no answer text. Reads stored runs only: it never starts a run and uses no quota. An id that is not a successful run of a tracked prompt in this project (another project's id, a failed run, a prompt removed from tracking) is listed in notFound instead of failing the call.

get_prompt_answer

Lists the chart notes of this project: dated notes people (or agents through create_annotation) pinned to the trend charts to record what they changed, e.g. "/ai-instructions page added". They show as markers on every trend chart of the project and in the white-label report, so a change in a metric can be read against the measure that probably caused it. Each note: id, date (YYYY-MM-DD, a UTC day), text, tags ([] = none), source ("manual" = added in the dashboard, "mcp" = added through this server) and author (display name, null when unknown). Newest first, up to 200 notes; total is the full count in the window. Without startDate/endDate every note is listed; either bound may be given alone.

list_annotations

Lists the prompt tags of this project. Tags belong to the project's ORGANIZATION and are shared by all its projects: for the editor role and above the list is every tag of the organization, so a tag can appear with promptCount 0 (no prompt of THIS project carries it yet; tag_prompts can still add it); for a viewer, and over a project API token, it is only the tags this project's prompts carry. Each tag: id, name, color, description (null = none set), promptCount (live prompts of this project carrying it). Also totalTags (tags listed), livePrompts (prompts of this project) and untaggedPromptCount (live prompts with no tag at all). With includePromptIds: true, each tag also lists its promptIds and the response lists untaggedPromptIds ([] = none), ordered like list_prompts; ids only, no prompt text, so it is the cheap way to find prompts missing a tag. Removed prompts are never counted. Requires a Pro or Enterprise plan on the project owner.

list_tags

Returns saved prompts for the project, each with its tags ([{id, name}]; [] = untagged; null = the project owner's plan has no prompt tags) and enriched with its latest run per AI model (full answer, brandMentioned, mentionPosition — order the brand was named in the answer, 1 = first, null = not mentioned — with mentionedBrandCount, citedUrl, searchTerms — the web searches the AI ran, the query fan-out —, visitedUrls — every page the AI actually opened, competitors included —, urlVisitRate, sentiment, mentionedBrands, mentionedBrandSentiments, runAt). Response includes `total` (project-wide count) and `returned` (paged count). Core tool for diagnosing AI-visibility gaps, e.g. which prompts mention the brand late.

list_prompts

Lists the Ayzeo projects available to you over MCP — projects you are a member of whose owner holds an active Pro/Enterprise (or trial) subscription — with your role (owner/admin/editor/viewer) and a dashboard link. Use the returned project id as the projectId argument for other tools. If the list is empty, the response carries a message explaining why. Read-only.

list_projects

Runs an ad-hoc prompt search across the configured AI search pipeline and returns the full answer text, cited URLs, and search results. Counts against the project's daily AI-visibility-check quota (plan-dependent).

run_ai_visibility_check

Triggers a fresh run of a saved prompt by ID. Returns the same rich payload as run_ai_visibility_check plus the promptId — enables before/after comparisons after the user implements a content change. Counts against the project's daily AI-visibility-check quota (plan-dependent).

run_prompt

Triggers a fresh full website analysis for the project domain. Optionally accepts a URL (must belong to the project domain).

run_website_analysis