Skip to content

features.md

Latest commit

 

History

History
235 lines (179 loc) · 11.2 KB

features.md

File metadata and controls

235 lines (179 loc) · 11.2 KB
Raw

Feature workflows

Questions mapped to the tools that answer them. For exact request/response schemas, hit runtime MCP tools/list or read openapi.json; the tool reference has every MCP name.

Contents

Start with discovery

Look before you interpret. Five reads tell you what's actually wired up:

  1. server_info — process policy, enabled surfaces, version metadata.
  2. accounts_list — the configured accounts, no secrets.
  3. sites_list — the resource boundaries you're allowed to touch.
  4. diagnostics — local config/credential presence, zero provider calls.
  5. provider_readiness — live access for one account.

Read tools stay discoverable even when a provider isn't configured, so your client always gets a stable schema. "Unavailable" is not "empty" — a missing credential is a different answer than a clean report.

Diagnose a Google traffic drop

Work outward from the aggregate to the individual pages:

  1. google_search_analytics_summary for the affected period.
  2. google_search_analytics_comparison against the period before it.
  3. google_search_analytics_drop_attribution to see which queries or pages ate the loss.
  4. google_search_analytics_trend and google_search_analytics_anomalies to pin the timing.
  5. google_url_inspection or google_url_inspection_batch on the pages that dropped.
  6. google_sitemap_get for submitted/discovered counts and current errors.

Search Console lags upstream. Rankrat reports what the provider says — it does not turn a not-yet-arrived recent row into proof you lost traffic. Don't confuse the two.

More Search Console reads:

  • google_search_analytics_query — typed dimensions/filters/rows;
  • google_search_analytics_dimension_report — fixed query/page/country/device/appearance/time dimensions;
  • google_search_analytics_page_performance — one page's report;
  • google_sites_list and google_site_get — upstream property state;
  • google_sitemaps_list — the property's sitemap inventory;
  • google_indexing_metadata — last Indexing API notification metadata.

That last one is a trap: Indexing notification metadata is not general crawl or index status. For ordinary pages, use URL Inspection and sitemap state.

Understand a GA4 property

google_analytics_account_inventory lists every account and property a discovery-enabled credential can see. Expect properties to sit under an account you didn't expect — the Analytics UI picks a default at creation time and nobody ever moves them.

google_analytics_data_streams maps a property to its web stream and G- measurement ID. A tag can be installed perfectly and still feed a different property; match the measurement ID before you declare it broken.

Reports:

Tool Purpose
google_analytics_report Typed custom historical dimensions/metrics
google_analytics_realtime_report Current activity
google_analytics_content_performance Page/content engagement
google_analytics_landing_page_performance Landing page performance
google_analytics_organic_search_landing_pages Search Console-linked organic landing metrics
google_analytics_traffic_source_performance Source/medium evidence
google_analytics_ecommerce_performance Item-level ecommerce evidence
google_analytics_audience_segments Audience segment metrics
google_analytics_user_behavior New versus returning engagement
google_analytics_conversion_funnel Closed ordered event-step funnel

Writable mode adds account and property renames. Names are cosmetic — the numeric IDs stay the boundary and the reporting identity. There is deliberately no GA4 delete. Google gives no API to create a GA4 account, so onboarding can only create a property inside an account that already exists.

Diagnose Bing search and crawl state

Traffic and reporting:

  • bing_traffic_trend, bing_traffic_anomalies, bing_traffic_comparison;
  • bing_query_performance and bing_page_performance;
  • bing_query_page_performance and bing_page_query_performance;
  • bing_brand_analysis, bing_ranking_buckets, bing_query_opportunities, bing_page_opportunities, bing_opportunity_matrix;
  • bing_query_cannibalization when several pages fight over one query.

Crawl and indexing:

  • bing_crawl_issues and bing_crawl_stats;
  • bing_feeds for sitemaps/feeds;
  • bing_url_information for one bounded child URL;
  • bing_url_submission_quota — check it before you spend writes.

Keywords and backlinks:

  • bing_keyword_statistics and bing_related_keywords;
  • bing_link_counts for paged inbound counts;
  • bing_backlink_intelligence for bounded page walking plus referring-domain and anchor summaries.

The opportunity, brand, bucket, and cannibalization reports are deterministic local math over Bing's own data. The labels are evidence — not buying advice, not a promise anything gets indexed.

Compare providers

Rankrat keeps each provider's numbers separate instead of pretending a Search Console click and a GA4 session measure the same thing. They don't:

  • ga4_search_console_comparison — organic landing observations next to Search Console observations;
  • search_engine_comparison — Google and Bing traffic evidence side by side;
  • search_engine_traffic_health — separate anomaly evidence per engine;
  • ga4_pagespeed_correlation — one GA4 content row next to one PageSpeed analysis.

Use these to find where the sources disagree — a Search Console click, a Bing click, a GA4 session, and a PageSpeed lab result each measure a different system.

Audit a site

site_audit crawls one configured HTTPS site through a DNS-pinned public fetcher and hands back page evidence, findings, remediation text, and a normalized score. It checks:

  • HTTP status and fetch failures;
  • robots and sitemap discovery;
  • titles, descriptions, canonicals, noindex, language, and H1 count;
  • image alt text and Open Graph metadata;
  • duplicate titles and non-HTML sitemap entries;
  • restricted Google Indexing structured-data eligibility;
  • truncation when page/issue/depth limits are hit.

The crawler follows normalized same-site links only. It rejects redirects, IP literals, private/special/mixed DNS answers, and out-of-bound paths — it doesn't wander off your site. Findings tell the caller what to change; Rankrat does not edit your CMS or repo for you.

Structured-data checks stand alone too:

  • schema_validate_json_ld — a local JSON-LD string;
  • schema_validate_html — a local HTML string;
  • schema_validate_url — a bounded public HTTPS fetch.

These validate Rankrat's supported Google Indexing eligibility rules — not every schema.org vocabulary, and not Google's full Rich Results suite.

Improve internal links and content

  • internal_link_graph crawls the bounded site and returns incoming/outgoing relationships.
  • orphan_page_report joins crawl, Search Console, and GA4 to surface URLs that have provider evidence but no crawled internal path.
  • internal_link_opportunities suggests missing same-site links, and only when normalized title/path tokens overlap.
  • content_opportunities joins Search Console, Bing, GA4, and crawl evidence into deterministic ranked opportunities.

No LLM in that loop. The suggestions are lexical — they don't invent semantic relationships out of thin air. Crawl truncation is reported, so you know when a graph or orphan result is partial instead of trusting a half-finished picture.

Inspect backlinks

bing_backlink_intelligence is the backlink evidence, from one configured Bing Webmaster site. It reports only what Bing exposes for that account. It does not scrape competitors, manufacture links, buy links, or send outreach — it's a reader, not a link farm.

Inspect performance

  • pagespeed_analyze — PageSpeed Insights lab analysis;
  • pagespeed_core_web_vitals — selected field metrics;
  • crux_history — up to the configured bounded history periods for an allowed URL or origin;
  • cloudflare_analytics — bounded hourly traffic/cache groups;
  • five local Lighthouse tools — an isolated Chromium audit plus per-category findings.

PageSpeed and CrUX are external Google services. Lighthouse is local and optional — read Lighthouse before you turn on its browser worker.

Manage tags, redirects, and Bing content

Start with google_tag_manager_accounts_list to find the signed-in user's Tag Manager accounts. From there the Tag Manager tools cover typed list, create, update, delete, workspace-version, and publication operations for containers, workspaces, tags, triggers, and variables. Bump to a release that adds a Tag Manager scope? Re-run rankrat auth-google first — the grant has to carry the current edit, delete, version-edit, and publish scopes or those writes just fail.

clarity_insights reads one bounded Microsoft Clarity project export. It's a read-only project token with a small daily upstream quota — use it to inspect evidence, not as a polling monitor you hammer every minute.

edge_redirects_list, edge_redirect_upsert, and edge_redirect_delete are provider-neutral names; Cloudflare is the adapter that ships today. Rankrat only lists and changes redirects marked as Rankrat-managed — it will not clobber your hand-written provider rules.

bing_content_submission_create fetches an allowed public HTML page itself and sends that freshly fetched body to Bing. It does not take caller-supplied HTML, headers, or a free-form destination URL — you don't get to inject content through it.

Ownership and onboarding

site_ownership_check reads Google/Bing and public DNS status without ever handing back provider verification tokens. site_ownership_verify creates only the exact provider-issued DNS proofs through a supported adapter, then redeems them after propagation. Site onboarding creates GA4/Search Console/Bing resources under the configured accounts, in normal writable mode. Read Ownership and onboarding before you run either write flow.

Monitoring and writes

Monitors persist bounded site-audit snapshots and issue lifecycle state. Everything else that writes: submits sitemaps/URLs, publishes eligible Indexing notifications, notifies IndexNow, renames GA4 resources, verifies ownership, remediates discovery, and runs typed Tag Manager, managed edge-redirect, Bing content-submission, and finite Cloudflare cache actions. Provider acceptance and indexing progress can lag after the call returns — read Monitoring and remediation for the async and partial-failure semantics before you enable writes.