Google Search Console

Add a property to this account (needs GSC_ALLOW_WRITE=1). Verification still happens in Search Console.

Brand vs non-brand traffic. Give the brand in every script it is searched in, comma-separated (e.g. "toyota, تویوتا, トヨタ"). Each term matches its spelling variants, as whole words; with level=loose a term of 4+ characters also matches when glued to its neighbours (toyotacamry).

Build a Search Console regex for a term that works in any script and matches its common spellings. Use the result as query_regex in other tools or paste it into the Search Console UI (regex filter). RE2's \b is ASCII-only, so this uses Unicode-safe boundaries; Arabic script gets ی/ي, ک/ك, ه/ة, ا/أ/إ/آ classes with optional vowel marks and half-spaces; digits match Persian, Arabic-Indic and full-width forms; Cyrillic е/ё; case-insensitive. loose=True also lets words run together. Accent and kana variants are not covered.

Compare a period with an earlier one; biggest movers first. compare_to: previous (the same number of days just before) | year_ago (the same calendar dates last year) | 52_weeks (364 days back, weekdays aligned) | custom (give previous_start and previous_end). dimension: query, page, country, device or searchAppearance. Queries are matched across spellings.

Pages (or queries) that lost or gained ≥ threshold of their clicks vs an earlier period. compare_to: previous | year_ago (same calendar dates) | 52_weeks (weekdays aligned). Decayed = dropped, rising = grew, lost = had clicks before and none now, new = the opposite.

Which recent days are final and which are still changing, for daily and hourly data.

Remove a sitemap from Search Console (needs GSC_ALLOW_WRITE=1). The file itself is untouched.

Queries where two or more of your pages compete as separate results, ranked by impressions going to the non-best page. Spelling variants of a query are merged first, so «خرید ماشین» on page A and «خريد ماشين» on page B is caught. min_share is the impression share a page needs to count as competing.

Auth status, scope, write access, history coverage and the list of tools. Call this first when unsure.

Details of one property (exact URL and permission level). Accepts loose input like example.com.

Details of one sitemap: submission/download dates, errors, warnings, per-type URL counts.

Compare two stored periods (e.g. this quarter vs the same quarter two years ago) — beyond the API's 16 months. dimension: query, page, country or device.

Query the local history for any date range, grouped by dimensions (queries merged across spellings). dimensions: query, page, country, device, date. query_contains matches the multilingual match key, so «ماشين» finds «ماشین» and 空调 works without spaces.

Run a read-only SELECT on the history database. Table `rows` (site, search_type, date, query, qkey, page, country, device, clicks, impressions, position); `sync_days`. qkey is the multilingual match key.

What the local history contains: per property and search type, first/last day, rows, final days.

Clicks/impressions/CTR/position per day, ISO week or month from the local history — any length of time.

Hourly data for the last 10 days — Search Console's 24-hour view, and the only way to get true LOCAL days. hours=24 reproduces the Performance report's 24-hour view: the most recent 24 hourly points, including preliminary ones. Otherwise the last `days` (max 10) are returned. Search Console's daily numbers are Pacific-Time days; with timezone= (e.g. Asia/Tehran, Asia/Tokyo) the hours are re-cut into local days (by=day) or listed per local hour (by=hour).

Problems only: which of the given URLs are not indexed or have structured-data failures, and why. Up to 50 URLs.

Full URL Inspection for one page: index verdict, crawl, canonical, robots, sitemaps, rich results, mobile.

Inspect up to 50 URLs in parallel (comma- or newline-separated). Quota is checked first. Columns: `index` is the indexing verdict, `rich results` the structured-data verdict — a page can PASS one and FAIL the other. URLs not reached within the time budget are marked and can be re-sent.

How many URL inspections this server has used today for a property (Google allows ~2,000/day).

Queries typed with the keyboard on the wrong layout (e.g. "ovdn lhadk" = «خرید ماشین» on a Persian keyboard). A hit is reported only when the remapped text is a query that really appears in the data, so the list is precise. include_unmatched=True also lists vowel-less Latin queries that map cleanly onto a layout (more findings, some false positives). layouts: any of fa, fa2 (the two Persian layouts in common use), ar, ru, he.

Share of clicks and impressions by the script/language of the query (Persian vs Arabic vs Latin…).

List every Search Console property this account can see, with the permission level.

Sitemaps submitted for a property, with errors/warnings and URL counts. Pass sitemap_index to list its children.

Query/page pairs on page one whose CTR is far below what their position should earn — title/snippet work. Expected CTR is the site's own median per position when there is enough data, otherwise a benchmark curve; ratio=0.5 flags rows under half the expected CTR.

Which pages rank for one query — in each of its spellings — and how the impressions split between them.

One-screen summary: totals, trend, top queries and pages, devices, countries, and how fresh the data is. granularity: day, week, month or auto — the same choice as the Performance report's time-granularity menu (for hourly, use hourly_performance). auto picks day up to 31 days, week up to six months, then month.

Which queries send traffic to one page. Spelling variants are merged unless group_variants=False. Give the exact page URL. If nothing matches exactly, pages whose URL contains the text are used instead and listed, so a path such as /blog/ works too.

Search Analytics rows with any dimensions and filters. The general-purpose query tool. Args: site_url: property (sc-domain:example.com, https://example.com/ or just example.com). days / start_date / end_date: range; explicit dates (YYYY-MM-DD) win over days. dimensions: comma list of query, page, country, device, date, searchAppearance, hour. search_type: web, image, video, news, discover, googleNews. query_filter: a term in ANY language; case-insensitive, and matches its common spellings (ی/ي, ک/ك, half-space, vowel marks, Persian/Arabic digits, е/ё). query_regex / query_regex_exclude: raw RE2 regex (see build_query_regex for a safe one). page_filter: substring of the page URL; page_exact: the exact URL; page_filter_exclude: substring to leave out; page_regex: RE2 regex on the URL. country: ISO-3166-1 alpha-3 (IRN, USA…); device: DESKTOP, MOBILE, TABLET. data_state: all (matches the UI, default) or final. group_variants: merge spelling variants of the same query (only when dimensions=query). level: standard or loose grouping (loose also merges spacing and accent differences). sort_by: clicks, impressions, ctr, position — applied to the fetched rows; Google itself always returns the top rows by clicks. limit: rows shown. max_rows: rows fetched.

Keywords that Search Console splits across several spellings, with their real combined totals. Groups queries by a per-script match key: Persian/Arabic letter forms (ی/ي, ک/ك, ه/ة, ا/أ/إ/آ), half-space, vowel marks, digit scripts, kana width, case, separator punctuation — also inside mixed queries such as «خريد iphone 13». level=loose additionally merges spacing, accents (café/cafe), hiragana/katakana, Simplified/Traditional Chinese (with the zh extra). Only groups with at least min_variants spellings are shown. source=history uses the local store.

Forget the cached client and sign in again (switch Google accounts or scopes).

Remove a property from this account (needs GSC_ALLOW_WRITE=1). Data is not deleted at Google.

Queries ranking just off page one (default positions 8–20) with real demand — the quickest wins. potential_clicks estimates extra clicks at position 5. Spelling variants are merged first.

Submit (or resubmit) a sitemap URL (needs GSC_ALLOW_WRITE=1).

Copy Search Analytics rows into the local SQLite history, one day at a time (keeps data past 16 months). Days already stored and final are skipped; recent or empty days are refreshed. dimensions defaults to query,page; add country,device for more detail (more rows). One call works for about 45 seconds and then reports how many days are left — call it again to continue, or run `gsc-mcp-full sync SITE --days N` in a terminal (no time limit; put it in cron to keep history growing).

Most demanded words across all queries — works for Chinese/Japanese/Thai (no spaces) too. Uses jieba / fugashi / pythainlp when installed (`pip install gsc-mcp-full[zh]` etc.), otherwise a script-aware fallback. Terms are merged across spellings.