MisarSEO MCP Server

Start monitoring one competitor page for changes: title/meta, a real line-level content diff (added/removed/unchanged line counts — never a fabricated similarity score), outbound links, and structured-data types. Checked once a day; the first snapshot is captured on the NEXT scheduled pass, not immediately. FREE — this codebase's own SSRF-safe fetch, no paid provider. Idempotent (re-adding an already-tracked URL is a no-op). A project can monitor at most 20 URLs.

Disavow file management: mark one URL or domain for disavowal. FREE (no credits), but this is a MUTATION. entryType 'url' disavows exactly that page (value must be a full absolute URL); entryType 'domain' disavows the whole domain AND its subdomains (value is a bare host, no scheme) — per Google's own documented disavow-file format, there is no way to disavow just a subpath.

Link Building Tool: add one prospect domain/page to a project's list. FREE (no credits), but this is a MUTATION. White-hat scope: contact fields are plain manually-typed free text — NEVER auto-scraped or guessed. New prospects start at status 'prospected'; use update_link_prospect_status to move them through the pipeline.

Link Building Tool: add one manual activity-log entry to a prospect, e.g. "emailed on 9/5" or "follow-up sent". FREE (no credits), but this is a MUTATION.

Adds projects to a portfolio. Returns `{ added }` — only ids that resolve to a project of the CALLER'S organization are added; any other id is silently dropped rather than erroring the whole call. FREE: database only, no SEO data provider is called and no credits are spent.

Adds a chart note to a rank tracker's own timeline. FREE: database only, no SEO data provider is called and no credits are spent. Config-level, not per-keyword or per-device. A config can hold at most 100 notes.

Adds keywords to an existing rank tracker config. Returns `{ added, addedIds }` — `added` counts only genuinely new keywords; anything already tracked for this config is silently skipped, so `added` can legitimately be lower than the number of keywords sent. FREE: database only, no SEO data provider is called and no credits are spent — but this CAN fail with a plan-limit error if adding these keywords would put the organization over its plan tier's rank-tracked-keywords cap, since that is a count limit, not a spend limit. This does not trigger a rank check or refresh keyword metrics — call get_rank_tracker afterward once you want positions.

Patches: AI-drafts title AND metaDescription for up to 10 pages in one call. This is a MUTATION and costs one AI-gateway operation per field per page (up to 20) — never a bulk discount. One page/field failing is reported in that entry's own error without sinking the rest of the batch. Requires an org admin.

Capture a domain-authority snapshot now: calls the same free resolver get_domain_overview already uses and records today's reading, so get_domain_authority_history has a trend to read back. FREE (no credits), but this is a MUTATION — idempotent per UTC day, so calling it twice in one day does not write twice. Useful for a domain get_domain_overview would not auto-capture on its own — e.g. a tracked competitor rather than the project's own domain.

Classify 1-200 keywords' search intent from their wording: informational, navigational, commercial, transactional, or unknown. Uses the SAME wording heuristic keyword research already applies as its free-tier fallback, made directly callable for a keyword list with no research/volume data yet. FREE and purely local: no upstream call, no billing, no market dependency. Navigational detection is limited to destination/account phrasing ("login", "official website", ...) that means the same thing regardless of brand — a brand-name navigational query with no such phrasing classifies as `unknown` rather than a guessed label.

Runs ONE prompt through the explore_ai_prompt pipeline and checks a brand plus its named competitors against the SAME real per-model answers for mentions and citations. Bounded scope, stated plainly: this is what one prompt's observed answers show for the named entities — never a comprehensive cross-prompt visibility score. `mentionedByCount` / `citedByCount` are raw counts out of `successfulModelCount`, never a percentage or ranking. `entities[0]` is always the target brand; the rest are the competitors in submitted order. A model that failed upstream reports `status: "error"` with `mentioned`/`cited: null` for every entity — null, not false, because that model was never checked. REQUIRES A PAID PLAN and CHARGES the same as explore_ai_prompt for the requested models; comparing more competitors against that one set of answers costs nothing extra.

Publishes a new "What's New" post on the linked Google Business Profile location. It goes LIVE and PUBLIC immediately — there is no draft state and no undo here. Show the user the exact text and get their explicit approval before calling. Only standard posts are supported (no event, offer or alert fields).

Creates a local pack rank check config: where businessName sits in Google's local pack for one keyword at one location. The location is composed into the query as "<keyword> in <locationLabel>" — not geocoded. Run checks with run_local_pack_rank_check and read them with get_local_pack_rank. FREE: database only.

Creates a geo-grid rank-tracking config: a 3x3 grid of points around a centre coordinate, checking ORGANIC rank for one keyword and domain from each point. centerLabel must name the place the grid is centred on — a heatmap with no stated centre is a claim about nowhere. Run it with run_local_rank_grid and read it with get_local_rank_grid. FREE: database only.

Creates a portfolio. Returns `{ id }`. Starts with no member projects — use add_portfolio_projects next. FREE: database only, no SEO data provider is called and no credits are spent.

Creates a rank tracker config (domain + market + devices + schedule) for a project and returns its configId. Next, add keywords with add_rank_tracking_keywords. Positions are collected by the SCHEDULED check ('daily' or 'weekly'; 'manual' means never on a schedule) and read with get_rank_tracker — nothing here triggers a check on demand. One tracker per (domain, market) per project: repeating a pair is a validation error, not a second tracker. FREE at creation: database only, no credits spent.

Creates a report definition with a chosen set of widgets, in print order. Returns `{ id }`. FREE: database only, no SEO data provider is called and no credits are spent.

Deletes a brand monitor, stopping its scheduled scans. Requires the admin role in the calling organization. Get the monitorId from list_brand_monitors.

Removes the owner's published reply from one Google review (reviewId from get_gbp_reviews). The removal is public and immediate. Confirm with the user before calling.

Deletes a local pack rank check config, INCLUDING its whole check history. Confirm with the user before calling.

Deletes a geo-grid rank-tracking config, INCLUDING its whole run and point history. Confirm with the user before calling.

Deletes a portfolio. Member projects themselves are untouched — only the grouping is removed. FREE: database only, no SEO data provider is called and no credits are spent.

Deletes a report definition. FREE: database only, no SEO data provider is called and no credits are spent.

Deletes a report definition's schedule. Requires the admin role in the calling organization, same as creating one. FREE: database only, no SEO data provider is called and no credits are spent.

Deletes a tracked prompt, stopping its scheduled checks. Requires the admin role in the calling organization. Get the trackedPromptId from list_tracked_prompts.

Patches: stops auto-publishing freshly AI-drafted patches for this issue type — future drafts wait for a manual publish_patch call again. Already-published patches are unaffected. FREE (no credits), but this is a MUTATION. Requires an org admin.

Full disconnect: drops BOTH the Rank Tracker and Site Audit bindings and the stored grant. Breaks Ahrefs data for every member of the organization, not just the caller — requires at least the admin role.

Remove a project's Baidu Webmaster connection. Breaks Baidu push submission for every member of the organization, not just the caller — requires at least the admin role.

Unlink a project's Bing Webmaster site. Breaks Bing Webmaster data for every member of the organization, not just the caller — requires at least the admin role.

Unlink a project's Google Analytics property. Breaks Analytics data for every member of the organization, not just the caller — requires at least the admin role.

Removes the project's link to its Google Business Profile location. Only MisarSEO's own record of the link is removed — the location on Google is untouched — but every GBP tool stops working for the project until it is linked again.

Unlink a project's Yandex Webmaster host. Breaks Yandex Webmaster data for every member of the organization, not just the caller — requires at least the admin role.

Patches: drafts a title/metaDescription/canonical/redirect fix for one page's finding. FREE for canonical/redirect (deterministic — canonical is always the page's own URL, redirect always needs an explicit value); title/metaDescription cost one AI-gateway operation UNLESS `value` is given, which sets an explicit replacement instead of drafting one. This is a MUTATION — it creates or overwrites this page+field's draft, but does not publish it (use publish_patch for that). Requires an org admin.

Patches: from now on, a freshly AI-drafted patch for this issue type publishes immediately instead of waiting for a manual publish_patch call — capped at 20 auto-publishes per project per day. A hand-typed replacement value (draft_patch's `value` argument) is NEVER auto-published regardless of this rule. FREE (no credits), but this is a MUTATION. Requires an org admin.

Runs one prompt through up to 4 model personas and returns each answer verbatim, with real citations when webSearch is on. The four persona labels (chat_gpt/claude/gemini/perplexity) are NOT vendor-verified — none is guaranteed to be that vendor's own model; 'claude' and 'perplexity' in particular have no real vendor route at all. Treat every answer as 'a model answering as X', never as ground truth from X. REQUIRES A PAID PLAN and CHARGES one shared AI-gateway operation per call (not per model) — plus one more if analyzeSentiment and highlightBrand are BOTH set.

Disavow file management: export a project's disavow entries as the exact .txt content Google's disavow tool accepts (support.google.com/webmasters/answer/2648487) — one bare URL per line to disavow that exact page, or `domain:<host>` to disavow an entire domain and its subdomains. FREE (no credits). Save `fileContent` to a .txt file as-is (UTF-8).

Link Building Tool: export a project's prospect list as CSV text. FREE (no credits). Save `csvContent` to a .csv file as-is (UTF-8).

Compares domains competing in search results for a supplied keyword set. Useful for market and search-intelligence reports; not radius-based local SEO. The inverse of get_keyword_gap, which asks what one competitor ranks for that we do not. CHARGES CREDITS ON THE PAID DATA PROVIDER, which is also the only one that honours `resultTypes` and `includeSubdomains` and that measures `visibility`, `etv` and `rating`. On the default free provider competitors are aggregated from one live SERP per keyword: NOTHING IS CHARGED, only ORGANIC results are counted whatever `resultTypes` says, `includeSubdomains` is ignored, and visibility/traffic-estimate/rating are left unmeasured. Sorting never treats an unmeasured field as zero: rows the provider did not measure sort last and are reported in `unmeasuredForSort`. If that count equals the row count, the list is NOT ranked by anything — say so instead of presenting the first row as the strongest competitor. Omit `locationCode`/`languageCode` to use the project's default market.

Compare a site audit against an earlier one: which issues are new, which were fixed, how many persist, and the health-score delta. Only pages BOTH crawls visited are compared, so a shorter crawl never reports the pages it missed as fixed — those are reported separately as notComparableCount. `newIssues` only ever covers an existing page newly showing a finding, never a brand-new page's own findings (those fold into notComparableCount/addedUrlCount instead, since a page neither crawl could compare has no meaningful 'new' vs. 'always there' answer). Returns an error when there is no earlier completed audit to compare against. Free — charges no credits.

Core Web Vitals report: LCP/INP/CLS per page/strategy this audit already ran Lighthouse against, classified against Google's own published thresholds. No new fetch — reads audit_lighthouse_results, the same table the results UI already renders as raw numbers. LAB DATA ONLY, never Search Console's real Chrome UX Report (CrUX) field data — every number is a single simulated Lighthouse run, not an aggregated real-user measurement. This codebase's audit pipeline runs local headless Chrome as its PRIMARY path, so CrUX field data is not reliably available for any given page. ALWAYS narrow on `state` first: `no_lighthouse_data` (this audit has no Lighthouse rows at all — an absence of data, NOT every page passing) or `analysed`. Errors when no audit with that id belongs to the project. `pages[].allGood` is true only when all three metrics were measured AND all three are "good". Free — charges no credits.

HTTPS Report: percentage of pages secure, mixed-content and insecure-link counts, homepage HTTPS-path status, and HTTP URLs in an HTTPS site's sitemap — for a project's latest completed crawl (or a named earlier one). Derives entirely from already-computed crawl data — no new work at read time. SOCKET-LEVEL TLS CHECKS STAY EXPLICITLY UNBUILT: expiring/expired certificate, old security protocol, incorrect certificate name, and subdomains without SNI/HSTS all need a real TLS handshake this fetch-based crawler does not make. `notDerivable` lists these with the reason. `homepageStatus.redirectsToHttps` is null (not measured), never assumed healthy, unless a standalone fetch actually observed a failure — this report cannot distinguish "checked and fine" from "never checked" from stored issue rows alone. ALWAYS narrow on `state` first: `no_audit` (no completed crawl yet), `analysed`. Free — charges no credits, makes no external call.

Internal Link Report: outbound/inbound internal link counts per page, orphan pages, dead ends, and click-depth distribution for a project's latest completed crawl (or a named earlier one). No new computation — reads the same audit_pages/audit_issues columns already powering the in-app internal-links page. INTERNAL LINKS ONLY. External links are explicitly out of scope: an external link graph needs data this codebase does not collect. `totals.externalLinkOccurrences` is the one external-facing number here — a raw per-page outbound tally, never a graph. ALWAYS narrow on `state` first: `no_audit` (no completed crawl yet), `signal_not_captured` (a crawl exists but its link counts predate the inbound-link/click-depth migration or were otherwise never computed — an absence of data, NOT a site with no internal links), `analysed`. `notDerivable` is returned with every state and lists what this report can never compute from what is stored, and why. Free — charges no credits, makes no external call.

Structured Data Report: pages grouped by schema.org @type, which pages have none, and which pages carry an unparseable JSON-LD block — for a project's latest completed crawl (or a named earlier one). Derives entirely from structured data already extracted at crawl time — no new crawl-time work. PRESENCE AND PARSE VALIDITY, NOT A RICH-RESULTS ELIGIBILITY CHECK. Only `typeGroups[].invalidCount` on the BreadcrumbList group is a real, verified per-type validity verdict; every other type reports only whether pages declare it and whether ANY block on the page failed to parse (page-level — never attributable to one type when a page declares several). ALWAYS narrow on `state` first: `no_audit` (no completed crawl yet), `analysed` (`typeGroups: []` means every page was measured and none declared any structured data — a real finding, not a withheld section). `notDerivable` lists what this report can never compute, and why. Free — charges no credits, makes no external call.

Anchor-text distribution for a domain or URL, one page at a time: which anchor texts link to it, with backlinks, referring domains, referring pages and dofollow links per anchor. This is what reveals an over-optimised profile (one exact-match commercial phrase dominating) or a hacked one (pharma/casino anchors nobody wrote) — both invisible link by link. CHARGES CREDITS ON THE PAID DATA PROVIDER (~200-500 typical). UNAVAILABLE ON THE DEFAULT FREE DATA PROVIDER, like the rest of the backlinks surface: SEO_DATA_UNAVAILABLE there means the question was never asked, never an empty distribution. There is deliberately no spam-score column — an averaged spam score is not a spam score.

Returns a backlinks profile summary (total backlinks, referring domains, referring pages, rank) plus the top referring domains by backlink count. CHARGES CREDITS ON THE PAID DATA PROVIDER (~200-500 typical): two lookups per call, the summary and one page of referring domains. UNAVAILABLE ON THE DEFAULT FREE DATA PROVIDER: a backlink graph needs a proprietary web-scale crawl index, so every method of this surface throws SEO_DATA_UNAVAILABLE there and nothing is charged. That error means the question was never asked — never report it as a site with no backlinks. Self-hosted deployments also need the Backlinks API enabled on their DataForSEO account. The two halves count DIFFERENT populations: the summary totals are unfiltered, while the referring-domain list has domains above the spam-score threshold filtered out. Do not present the listed domains as the top slice of the summary total, and do not subtract one from the other.

The project's current Baidu Webmaster (百度搜索资源平台) connection, plus today's push-quota usage. `today.lastRemain` is Baidu's own authoritative remaining-quota count as of the last push (null until the first push of the day) — `today.assumedDailyQuota` is a courtesy default only, never a real ceiling, since Baidu computes quota per-site dynamically and publishes no fixed number.

GetLinkCounts: own-site inbound-link page list for a project's matched Bing site — the focused, single-call form of the backlinks field already embedded in get_bing_site_overview. FREE (no credits). OWN-SITE ONLY: Bing exposes no competitor/any-domain backlink lookup, and some verified sites are known to return an empty result here even when populated. Bring-your-own-key: requires BING_WEBMASTER_API_KEY. Answers { ok: false, reason: 'not_configured' } when unset, or 'no_matching_site' when the project's domain has no verified match on the account.

GetQueryTrafficStats + GetPageStats + GetPageQueryStats + GetQueryPageStats + GetQueryPageDetailStats: per-query and per-page search-performance detail for a project's matched Bing site. FREE (no credits). pageStats (site-wide, weekly) is always attempted. queryTrafficStats/queryPageStats are null unless `query` is given; pageQueryStats is null unless `pageUrl` is given; queryPageDetailStats is null unless both are given — a null section absent from `errors` was not requested, not a failure. pageStats carries one unresolved caveat: Microsoft's own doc page is internally inconsistent about whether that method's `query` field on each row holds a query string or a page URL — surfaced as-is. Bring-your-own-key: requires BING_WEBMASTER_API_KEY. Answers { ok: false, reason: 'not_configured' } when unset, or 'no_matching_site' when the project's domain has no verified match on the account.

GetFeedDetails: the same Feed shape as one entry from Bing's GetFeeds list, for one specific feed URL on a project's matched site. FREE (no credits), read-only. Bring-your-own-key: requires BING_WEBMASTER_API_KEY. Answers { ok: false, reason: 'not_configured' } when unset, or 'no_matching_site' when the project's domain has no verified match on the account.

GetKeyword + GetKeywordStats + GetRelatedKeywords: real Bing impression data for a query, its recent weekly history, and related queries — read straight from the Bing Webmaster API. FREE (no credits). Unlike derived/estimated keyword metrics, these numbers come from Bing's own search-impression logs. Bring-your-own-key: requires BING_WEBMASTER_API_KEY. Answers { ok: false, reason: 'not_configured' } when unset, or 'no_matching_site' when the project's domain has no verified match on the account.