MCP tools

Shows where one figure comes from: the stored tables, the exact window and compare window, data_through and complete_through, which days are provisional, the sync runs that wrote them, how much traffic sits in query rows, and the Search Console filters that reproduce it.

Lists what SerpUp did for the workspace, newest first: syncs, audit scans, anomaly scans, investigations, opportunity refreshes, fix packs and verifications, each with outcome (running, ok, partial, skipped, failed), counts, times and the reason for a skip.

Totals of SerpUp's work for the workspace in the last 24 hours, 7 days and this month (syncs, checks, anomalies, investigations, fix packs, verified results, skipped steps) and the plan's daily audit allowance used; every figure is a sum of recorded events.

No arguments.

Newest alert events of the workspace or one property: what fired and when, with severity, title and for each event the send log per endpoint (queued, sent, failed or suppressed, with the reason) or no_endpoint_reachable when nothing could be sent.

A ready-to-relay report of one property or the workspace: headline, what changed against the previous 28 days (with data_through and provisional days), why (evidence, other explanations, plain-word confidence), what to do next, what is done with its verified or unverified outcome, limits and the 7, 14 and 28 day follow-up, plus markdown and compact_markdown.

The Dashboard bands in one call: incidents first (visibility collapse, confirmed historical indexation decline; onset, levels, detection confidence, causal chain with cause not verified, state, next_tool; warnings hold a current high non-indexed share), then health, what_changed, fix_next, being_fixed, results.

The best next steps in lanes, each item tagged with its queue (incidents, technical_fixes, growth_opportunities, monitoring): critical (a visibility collapse, a confirmed historical indexation decline or a severity 2 or 3 drop, each with cause_status; always first), search (findings and briefs by gain range times confidence over effort), site_checks, in_flight (tasks), setup, plus next_moves.

Freshness and sync state of one property: when its data runs through, whether syncing is paused, the sampling note, plan context and whether Search Console gives SerpUp access to the site.

The stored Site health score (0-100) and its six weighted parts (indexing and crawl, speed, on-page, trust, content, schema) with state, score and unverified counts, the change since the last snapshot and optional history.

Pages the whole open backlog for the connected properties, best first: findings, open investigations, tasks and content briefs, ranked by gain range times calibrated confidence over effort (never a point gain).

Stores one verdict, done or not_useful, on an insight from get_next_actions, against the person the connection acts for; a second call with the other verdict changes it.

Ranks the property's SEO opportunities by expected gain, each as a range with a calibrated confidence tier, severity and effort; among comparable gains of the same tier, revise goes before merge before create (a tie-break, not a score).

The machine-actionable fix for one audit finding or task: ordered steps with a repository path and anchor, a diff only for a new file, snippets, config, redirect map or JSON-LD, pre-checks, rollback and the post-check.

The re-check results of an applied fix: the on-demand check and the windows at 7, 14 and 28 days, each with its verdict (improved, no change, regressed, inconclusive), the reason, the audit check before and after, and the measured search impact with its limits.

Report that the fix was applied and deployed, with evidence (pull request, commit, live URL, changed files).

Re-checks every task reported applied and still waiting for a verdict, in one call, with the same service as verify_task; it never approves, verifies as a person or closes a task.

Re-run the audit check behind a task against the live site now and compare it with the check as it was before the fix.

Creates a task that freezes its evidence, acceptance criteria and verification check, from a finding (finding_id), a score_opportunities row (opportunity_id), a failing audit check (audit_check), an investigation hypothesis or by hand.

Turns up to 20 findings or failing audit checks into tracked tasks in one call, each exactly as create_task makes it.

One task in full: status, the evidence it froze when it was made (metric, gain range, confidence, URLs, facts), acceptance criteria, the verification check, instructions, blockers, deploy evidence and, once applied, the verified, unverified or regressed outcome.

Lists tasks newest-updated first with their status, priority, source finding or audit check, acceptance progress and the exact next call, filtered by status, kind, assignee, source, due_before, finding_id, check_id or text.

Moves a task to in_progress or deployed (deployed needs evidence_url or attestation: true), or adds a comment with note.

Lists queries served by two or more of your pages, split into competing groups (pages swap places), coexisting groups and unverified groups (a check could not run, for example pages not crawled yet), each with its safeguards.

Trend, position distribution, ranking pages, seasonality and related queries for one query or a regex group.

Labels one month of Search Console queries by shape with SerpUp rule-based heuristics (bare reply, pivot follow-up, conversational question, AI tracker probe, agent instruction, pasted text, conventional) and returns queries, impressions, clicks, zero-click share and mean position per shape, AI-shaped totals and top examples.

Compares your crawled pages with one competitor sample: click depth, internal links, structured data, title terms.

The last result of a saved site migration: clicks per day before and after go-live, recovery, the rule-based risk score and the old pages ranked worst first.

Finds days when clicks or impressions broke from their normal pattern, normalised for the day of week, with the size of each jump.

Finds dates when the level of clicks or impressions moved to a new normal and stayed there, with the before and after level.

Finds pages or queries that lost traffic steadily over months against their own baseline, with clicks lost per month and a note when the drop looks seasonal.

Finds striking-distance pages (average positions 4 to 20 by default) that rank close to the first results, with the gain read off the CTR curve named in curveSource.

Projects clicks or impressions forward from stored history with a seasonal model, giving a central line and 80 and 95 percent bands plus a backtest error.

Five separate AI signals in `signals` (AI-prone query estimate, Google AI features, assistant sessions, AI conversions, crawler access): each has status, value or N/A, reason_code, reason, source and data_through.

Reads the stored weekly Chrome UX Report history for the site, a page or a saved origin: 75th-percentile LCP, INP, CLS, FCP and TTFB with good, needs-improvement and poor shares.

Queries or pages whose click-through rate is below (or above) the property curve at the same position, with the clicks missed.

One experiment of a property with its stored result and uncertainty (per group effect, interval and verdict as the Experiments page shows them), the URL groups it covers, and for a concluded one its frozen impact record.

One dated list of what SerpUp recorded around a traffic drop: change point, robots.txt, status, redirect, canonical and meta robots changes, deploys, audits, indexation first seen by inspection, notes, alerts, each with source, date confidence and evidence.

One investigation: the change, hypotheses with confidence wording (never confirmed below 0.9 calibrated), evidence basis, alternatives and `contributors`: the click change split by impressions, CTR and position and by country, device, brand, page group, locale, page and query, shares summing to 100% with an `unattributed` remainder, worded likely contributor, never cause.

Groups the site's own Search Console queries that pair a service with a place, near me or a postal code into service, brand and place, then flags a place ranked by a page that does not name it and a place with no page.

Services by cities for a local business: mapped page, weighted position, clicks, leads and NAP status per cell, plus gap cells with no page.

Splits the site's pages by clicks into head, chunky middle and long tail using the elbows of the click curve, with page counts, click shares and concentration.

The stored monthly trend of query shapes (heuristic labels, see analyze_query_shapes): per month the impressions and queries of every shape, the AI-shaped share of impressions and the rules versions used.

Keywords whose average position improved or worsened between two periods, with click and impression impact.

Per property, the percentage of clicks hidden by Search Console sampling at query and page level (query over 50 percent and page over 15 percent are High).

Groups known URLs by path segment with active and dormant counts, share of URLs against share of clicks (a heuristic about focus) and a rules-based decision per segment (revise, merge or redirect, hold, no change, not enough data), plus queries without a page and queries split across page pairs with a verdict.

Tracked keywords with the latest SERP vendor sample (a third-party estimate at one location and device), the previous sample, change and ranking URL, and apart the Search Console average position.

Whether impressions or clicks fell 60% or more for 3 or more complete days (from 14 days of history): onset, before and after levels, detection confidence, and the triage (site-wide or few pages, index verdicts of lost pages, update overlap, manual action step for the owner, three moves).

Ranked internal link suggestions: source, destination, anchor, evidence strength and the queries behind each (read only).

Deploys, crawl changes, deployed tasks, data gaps and incidents recorded for a property in a date range, plus the onset day of an active visibility collapse as an incident row when it falls in the range (derived from the daily series, not a recorded change; a range that starts after the onset shows none, so read get_visibility_collapse for the date).

The SEO experiments of one property (Pro plan): name, hypothesis, kind (compare or causal), status, ranges, change date, whether a result is stored and when it was concluded.

Queries or pages that appeared or disappeared between two periods, with the counts of all four kinds.

Starts an investigation of a finding or a question over the last 28 days of one property; asking again about the same finding returns the open one; uses one of the monthly investigations.

Template clusters with indexed ratio and interval, duplication, thin share, depth, worst finding, change since the last run and tasks created (read only).

Runs a custom question on the stored Search Console data: dimensions (date, query, page, country, device, searchAppearance), metrics, filters, dates and an optional comparison.

Structured data problems by rule with page counts, plus `reconciliation`: `coverage` (rules checked, audit date, what is not checked, coverage_gaps) and Google against SerpUp per finding (coverage_gap, serpup_only, agree, needs_verification).

Confirmed technical changes found by the monitor, counted by severity (read only).

The 21 Merchant feed checks with the number of errors, warnings and suggestions found by the latest finished audit (read only).

AdSense readiness checklist of the newest run: blocking and advisory checks with status, pages affected and the source of each rule.

Agent readiness from the last finished check: per AI crawler the robots.txt answer (allowed, blocked, not mentioned) with the rule line and its role, plus llms.txt (informational, not scored), raw-HTML text, Markdown and /.well-known.

Status, progress and, once complete, the scores of one audit run: pass rate, per group, per E-E-A-T axis, and the unverified (each with a reason) and n/a counts.

One check of a run with its full evidence, its basis (specification, Google-recommended, best practice, optional, heuristic) and confidence and, when it needs action, a fix pack: where the change goes, copy-ready snippets and a verify rule.

Latest finished PageSpeed Insights test of one page: Lighthouse score, lab metrics, Chrome UX field data (page, else origin, else none), budget verdicts, top opportunities and check counts.

Trust checklist from stored page snapshots: per policy page and page element, the status present, missing, unknown or not applicable, with evidence.

Check results of a completed run, failing checks first by priority: check_id, title, outcome (pass, action, unverified), presentation (finding, suggestion, observation), basis, confidence, severity, effort, one evidence line and, for each action, a make_task call.

Starts an audit scan of one site (kind eeat or crawl, or speed for one page); a scan of the same kind already running is returned instead of a second.

Queues PageSpeed Insights tests for several pages of one site (your urls, else pages picked across the site sections).

How many synced Merchant products break each title rule: short, long, capitals, promotion, symbols, missing brand, odd spaces, repeated word, duplicate title, no colour or size (read only).

Waits up to timeout_seconds (at most 25) for audit, speed or on-page runs to finish and returns each run state: status, percent (null and labelled indeterminate when the worker reports no count), last_message, eta and a done flag.

Referring domains per backlink source (Bing, a Search Console Links import, connected vendors): new, lost and changed between the newest snapshot and the previous.

Creates a rolling client report of one site and invites the listed clients by e-mail (the report renders in the background; plan limits apply).

Deletes one shared report and every client link to it; clients lose access at once and it cannot be undone.

Builds a report on demand: re-renders one shared report (report_id) or runs one schedule now (schedule_id) and delivers it to that schedule's channels.

Affiliate links found on the property's pages: network, rel attributes, whether a disclosure phrase was found (a text heuristic) and destination health (blocked is unknown, never dead).

A content brief as JSON or Markdown with the limits of its evidence; pass brief_id, or topic_id (the opportunity id) for the newest.

Reads the scheduled workspace briefing: on or off, weekday, hour and time zone, where it goes (the owner e-mail, alert endpoints), the next send, and when it last went out with its outcome (sent, cannot_assess, skipped_no_moves, failed).

No arguments.

Clicks, impressions, CTR and position summed over all your properties for a stored global run, with the per-property breakdown, top countries and properties that failed to load.

How often SerpUp recommendations worked here, per task kind: worked of closed, median lift range, time to effect, sample size and inconclusive count; no median below three closed tasks.

The people reports are shared with, one row per address across all shared reports: how many reports each can open, total views, last viewed and the soonest access expiry.

No arguments.

Scheduled reports of the workspace (template, scope, cadence, hour, time zone, channels, active, next and last run, consecutive failures) and the latest 25 runs with status, period, data_through and which files exist.

The reports a workspace shares with clients: title, report type, site, status, render state, data_through, whether final and client count; with report_id also its settings and each active client (address, expiry, e-mail status, views).

Changes who can open one shared report: set the full client list (new addresses are e-mailed a link, dropped ones lose access), or revoke, renew or resend the link of one client.

Turns the scheduled workspace briefing (next 3 moves across all sites, once a week) on or off and sets weekday, hour, time zone, owner e-mail and alert endpoints; needs an owner.

Queues a background URL Inspection job for up to 500 URLs of one property, capped by the inspection quota left today (URLs over it are not queued and are listed).

Runs the Index Tracker now for the named tracked URLs (page_ids) or for what is due; uses URL Inspection quota, capped to what is left today, once every 10 minutes per person and property, and refused while the tracker is paused.

Crawler hits from the property's server logs per day, crawler family and trust level, with status classes.

Known URLs sorted into indexation states as Google last reported them, each with a count, exact URLs, the verdict date, freshness (fresh, aging, stale), last impression date and one next step; coverage says if enough URLs were inspected for a site-wide claim.

Pages that had high impressions before a drop (28 days before the onset, else the earlier 28 days) and that Google last reported as crawled-not-indexed, excluded or unknown, ranked by what they earned, with critical for top-click pages or 2%+ of impressions.

The persistent Index Tracker of one property: tracked URLs with confirmed status, risk and last check (section urls), or one tracker report (overview health score, trend, lost, freshness, states, coverage).

IndexNow state of one property (key state none, unverified, verified or failing, key file location, last verification error) and its submission history: when, source, engines, URL count, HTTP statuses and URLs skipped as recent.

The URL Inspection budget of one property today: used, remaining, remaining for bulk work, the ceiling SerpUp stops at (Google allows 2,000 a day, 100 stay free for Search Console), the split by source and when the Pacific-day count resets.

Asks Google, through the URL Inspection API, about one page of a property; uses one inspection of today's quota and queues a job the worker runs.

The sitemaps Search Console lists for one property from the latest stored snapshot: kind, status (ok, warnings, errors, pending), last submitted and read, errors, warnings and URLs discovered.

Stored URL Inspection results of one property as Google answered them, each with its date, age in days and freshness (fresh, aging, stale): the latest per URL, the history of one url, or the per-URL results of one job_id.

Clicks, impressions and pages without traffic for each sitemap of the newest analysis (read only).

Submits up to 100 URLs of one property to IndexNow so Bing, Yandex and others re-crawl them; URLs must be on the property host.

Adds or resubmits sitemaps of one property in Search Console after a preflight fetch; needs Full access to the Search Console property and queues a run the worker performs.

Adds a note to a property's timeline, for example a launch or migration; the same date and text within a minute create one note.

Search Console clicks and GA4 sessions of the same pages joined on the normalised URL (match both, gsc_only or ga4_only); ga4filters is the screen JSON.

GA4 sessions, engagement, key events and revenue by default channel group (read only).

GA4 sessions by country, with full country names (read only).

Event counts per value of one synced GA4 custom dimension (read only); pass dim as customEvent:name or customUser:name.

GA4 item views, add to cart, purchases and revenue per item (read only).

Pages whose title or path looks like a 404, with views, previous period and Search Console clicks (read only).

The events marked as key events in GA4: how often they fired and what they were worth (read only).

GA4 landing sessions, engagement, organic sessions, key events and revenue per landing page (read only).

Sessions, engaged sessions, key events and revenue of the site by day, from Google Analytics 4 (read only).

GA4 page views per page with the Search Console clicks of the same page (read only).

GA4 sessions by source and medium (read only).

Bing's crawl stats per day (pages crawled, in index, crawl errors, inbound links, HTTP codes, robots blocks) and the crawl issues it lists per URL, from SerpUp's stored sync, with open issues by type.

Bing Webmaster Tools clicks, impressions, CTR and position from SerpUp's stored sync: site totals by day, top queries or top pages, with a period summary and an optional comparison.

Clicks and impressions split into branded, non-branded and unattributed queries, with the previous period.

Click-through rate of all your properties by Google position 1 to 20, with the gap to each verified benchmark.

Health of the synced Merchant Center product feed: products by listing status (approved, pending, disapproved, demoted) and availability, the issues affecting the most products, destination status, the mapped account, the last connection test and sync.

Free listing clicks, impressions and CTR per product from Google Merchant Center (paid ads excluded), with the change against the previous period.

Clicks, impressions, CTR and position for a property over a range, with the previous period.

Queries and impressions in the position buckets 1-3, 4-10, 11-20 and 21+, with the previous period.

Distinct queries per day, in total and by position bucket.

Whether Bing Webmaster Tools and Yandex Webmaster are connected to a property: linked site, connection state, last successful sync, how far each stored dataset runs and which tool reads it, plus the provider data SerpUp does not store.

Pages with clicks, impressions, CTR and position, and the change against the previous period.

Search queries with clicks, impressions, CTR and position, and the change against the previous period.

Yandex's index of the site from SerpUp's stored sync: pages in search, downloaded and excluded per day, the site quality index (SQI) and the problems Yandex lists for the site.

Yandex Webmaster clicks and shows from SerpUp's stored sync: site totals by day, or top queries from Yandex's weekly lists with show and click positions, with a period summary and an optional comparison.

Notes people put on a property's timeline: launches, migrations, algorithm updates.

Clicks, impressions, CTR and position by country, with the change against the previous period.

Clicks, impressions, CTR and position by device, with shares and the change against the previous period.

Clicks, impressions, CTR, impression-weighted position and share of site impressions for the Merchant listings and Product snippets search appearances, with the change against the previous period.

Products of the synced Merchant Center feed with their listing status (approved, pending, disapproved, demoted), price, availability, destination status and the issues Merchant Center reports, worst status first.

The saved filters of one property that the key can see: name, scope (private or shared on the property), the encoded filter `f` the app applies to reports, and the per-property maximum.

Clicks, impressions, CTR and position by search appearance type.

Topic clusters of a property with their keywords and, when built, clicks and impressions.

Creates, updates or deletes one saved filter of a property; deleting cannot be undone and a shared (property) filter changes the view for everyone with access.

Creates, updates or deletes a topic cluster of a property; delete needs confirm: true, sent only after the person approved it, and plan limits apply.

Adds a place alerts are sent to (webhook, Slack, Teams or email); needs confirm: true and an owner of the workspace, and the plan limits the channels and the count.

Adds any public website as an audit-only site: audits and site checks work, Search Console and GA4 reports do not exist for it and say so, and ownership is not verified.

Puts a Search Console property that a connected Google account can already access on the dashboard; fails with plan_limit_reached when the plan is full.

Start here: lists the four jobs an agent is usually given (quick insights, quick fixes, make tasks, complete tasks) as the exact tool sequence for each, then every tool grouped by workflow with the scope each needs.

Text of a SerpUp lesson by slug: title, reading time, Markdown body and the glossary terms it uses.

Shows the workspace plan, the API rate limit per minute and hour with what is already used, and the other plan meters with their usage.

No arguments.

How far this workspace is through setup (Google connected, site added, first sync, audits run, task created, agent connected, report saved).

No arguments.

Lists where alerts are delivered (email, webhook, Slack, Teams): kind, name, whether it is active, why it was switched off, the failure streak and the last delivery.

No arguments.

Lists the alert rules (kind, site, severity, cooldown, schedule, last fired) and whether each is deliverable; a rule with no active endpoint is deliverable: false and fires into nothing.

No arguments.

Lists the properties (sites) this connection can read, with id, sync state and the date the data runs through; a site added for audits only has audit_only: true, no search data and an availability that says so.

Pauses, resumes or removes one alert endpoint (webhook, Slack, Teams or email); needs confirm: true and an owner of the workspace.

Asks Google again, now, which sites the connected Google accounts can see, after the owner fixed access in Search Console; once per 2 minutes per workspace.

Sends one clearly labelled test event ("Test event", kind webhook_test) to one endpoint, only with confirm: true and an owner, at most once a minute per endpoint.

Asks for a fresh Search Console pull of one property; allowed once every 15 minutes per property and it does not wait.

Replaces the tags of one property with exactly the names given; missing tags are created and an empty list removes all.