MCP App Store
by KeenEthics
GMB Club icon

GMB Club MCP Server for ChatGPT

By GMB Club

Overview

GMB Club connects ChatGPT to your Google Business Profile and local marketing stack. Read and reply to Google reviews with AI-suggested replies, schedule and publish posts to Google, Facebook, Instagram, LinkedIn and more, create and publish articles to your website, track SEO keywords and Search Console performance, monitor your local visibility, and send review requests to your customers. Everything is scoped to the business locations you explicitly authorize, and every action that publishes or sends something runs as a preview first and requires your confirmation.

How to connect

  1. ChatGPT: Open the GMB Club listing in the ChatGPT app directory and choose Connect. ChatGPT listing GMB Club exposes 40 tools here.

Tools

add_tracked_keyword

ChatGPT
Use when the user wants to track a new SEO keyword. Add a keyword to the tracking list for a bulle. The keyword will be checked monthly automatically. Returns the created tracked_id.

apply_gmb_optimizer_suggestions

ChatGPT
Use when the user wants to apply the optimizer's suggested improvements to their Google listing. Apply one or more accepted suggestions to a Google Business Profile listing. Each change is a {field, value} dict (field = description, categories, regularHours, serviceItems, attributes, etc.; same schema as the suggestions returned by get_gmb_optimizer(kind='audit')). For services and categories, pass the suggestion's items as they are (service lines {name, description, serviceTypeId?, evidence?}; category names or {name, displayName}): they are MERGED into the listing's current list, nothing already there is removed. A service line matching no current service is added only with a serviceTypeId (Google's predefined service) or an evidence; for a service the user explicitly asked for, pass evidence={'source': 'user_request'}. A line whose words match a current service (plural and word order ignored) updates that service's description instead of creating a twin. A line too close to an existing service (more than half of its words in common) is NEVER added, even with an evidence: it is listed in services_too_close [{nom, existant}]. Lines with neither match, serviceTypeId nor evidence are listed in services_not_added. The result also carries services_updated and services_added (names): report exactly these to the user, never claim more. Only a complete list in Google's format replaces it, removals included (serviceItems with structuredServiceItem/freeFormServiceItem; categories as {additionalCategories: [...]}, the primary category is never changed). Without confirm=true, returns a preview of what would be changed. With confirm=true, applies them via the Google API and invalidates the audit cache so a fresh audit can be run.

create_article

ChatGPT
Primary tool to write a blog/SEO article. Use when the user asks to write, draft or prepare an article for their website. Create an article on ONE or SEVERAL connected sites at once. Triggers AI content generation in the background by default. TARGETS : every connected platform is supported. Use prestashop_site_id / prestashop_site_ids[] (PrestaShop), shopify_site_id / shopify_site_ids[] (Shopify), wix_site_id / wix_site_ids[] (Wix), webflow_site_id / webflow_site_ids[] (Webflow), site_id / site_ids[] (WordPress). Lists can be combined (every article is identical, just pushed to N destinations, ideal for franchise networks publishing the same content on every establishment's site). Call list_sites first to get the right id and platform. When N targets are provided, N articles are created (one per site, trackable separately). Each site must be accessible to your token. Sites that fail tier gating are skipped. NO SITE CONNECTED : when this tool returns error='no_site_connected', tell the user in plain words that no website is linked to GMB Club yet, so their articles cannot appear online, then walk them through steps. Never describe such an article as published or online. Image options : pass featured_image_url with a public HTTPS URL (typically generated first via generate_post_image, or uploaded via upload_image, which accepts either base64 or HTTPS url). OR set generate_image to true to let the AI create an illustrative image automatically. When multiple sites are targeted, the same featured_image_url is applied to every article. Content : if generate_content is true (default), the AI writes the article in the chosen tone & language (~1-3 minutes). The article is then ready to publish: call publish_article afterwards to push to the destination. If generate_content is false, NO text is written : the article stays a draft (even after the optional image is generated) and CANNOT be published until you provide the body via update_article(content_html). tone accepts both English and French variants : 'casual' / 'décontracté' / 'amical', 'professional' / 'professionnel' / 'neutre', 'expert' / 'technique' / 'spécialiste'. Default 'professional'. language accepts a language code ('fr', 'en', 'es', 'de', 'it', 'pt'...) or a language name ('français', 'english', 'spanish'...). Leave it EMPTY to use the location's configured language (recommended unless the user explicitly asks for a specific language). This tool creates the article immediately (no confirmation step) and triggers the AI generation in the background. The widget displayed to the user will refresh on its own once the content is ready. Articles can be deleted or edited later if needed. When summarizing results to the user, refer to the article by its title, never by numeric ID.

create_qr_code

ChatGPT
Use when the user wants a QR code that sends customers to their Google review page. Create a new tracked QR code for a sphere (or specific bulle), with a target URL (Google review form, website, custom landing, etc). Returns the generated PNG URL and the tracked short URL. The QR is generated black-on-white by default ; pass color_hex (e.g. '#E2611B' for orange) to customize. By default the sphere logo is embedded in the center of the QR (with_logo=true) ; pass with_logo=false for a plain QR. Without confirm=true, returns a preview without creating anything.

create_scheduled_post

ChatGPT
Primary tool to create social media posts. Use when the user asks to post, publish or schedule content on Google, Facebook, Instagram, LinkedIn, TikTok, YouTube, Pinterest or Snapchat. Schedule a social media post on ONE or SEVERAL establishments at once. Requires content, platforms[] (facebook/instagram/linkedin/pinterest/tiktok/snapchat/gmb) and scheduled_at (ISO 8601). TARGETS : pass bulle_ids[] to publish on multiple establishments at once (essential for franchise networks: one prompt publishes to all their establishments, even across different spheres). For a single establishment, pass bulle_id. For a single sphere with exactly one establishment, pass sphere_id (auto-resolves the bulle). When N establishments are targeted, N posts are created (one per establishment, trackable separately). Each establishment must be accessible to your token. Establishments that fail tier gating are skipped (reported in the response). TIMEZONE : if scheduled_at has no timezone info (e.g. '2026-05-27T11:00:00'), it is interpreted in the LOCAL timezone of the first target establishment (typically Europe/Paris). To force UTC, append 'Z' (e.g. '2026-05-27T11:00:00Z'). When targets span multiple timezones, prefer Z-suffix to avoid ambiguity. IMAGES : image_url + image_mode='upload'. image_url can come from generate_post_image, upload_image (which accepts either an image_base64 attached by the user or a public HTTPS url), or any HTTPS URL. CAROUSEL (multi-images) : pass image_urls=[url1, url2, ...] (2 to 10 images, order = slide order) instead of image_url. Each URL can come from generate_post_image (kind='carousel' builds all the slides at once : pass its carousel object here too, so the user can edit each slide's text in the app), upload_image or any HTTPS URL. Instagram/Facebook publish a real carousel, LinkedIn a swipeable document (slides combined into a PDF), Pinterest max 5 images. VIDEO : pass video_url = a public HTTPS link to an MP4/MOV/AVI/WebM file (max 100 MB). The video is downloaded and hosted by GMB Club, then published natively (Reels, LinkedIn video, TikTok...). Video takes priority over any image. ONE POST PER MESSAGE : the same message on several networks is ONE call with all the platforms, never one post per network (the client's website lists each post, so it would show the message several times). Do not prepare one video per network either : GMB Club adapts the video format to each network itself. A post with the same text already scheduled for a location is reported in duplicate_warnings : offer to update it instead of creating a second one. STORIES : pass formats={'instagram': 'story', 'facebook': 'story'} to publish a story instead of a feed post (per platform, others stay posts). A story takes ONE image or ONE video of 60 s max, never a carousel or text only. Its caption is NOT shown : put the message INSIDE the image (generate_post_image with aspect_ratio='9:16' and text_elements), content may then be empty. No stickers (link, poll, countdown) : Instagram does not allow them from apps, say so plainly. Mention accounts in an Instagram story with instagram_mentions=['username'] (no '@'; they are notified). Snapchat formats : story (default), saved_story, spotlight (video only). MEDIA PER PLATFORM (same rule as the GMB Club composer) : Google Business takes ONE photo only, never a video (Google refuses it from apps; a video can only be added by hand on Google) ; a carousel is fine, Google Business receives its first image. Snapchat and YouTube take no carousel, YouTube takes video only, Instagram/Pinterest/TikTok/Snapchat/YouTube need a medium. A platform that cannot take the medium is LEFT OUT of the post (listed in platforms_excluded and media_warnings) while the others are scheduled : always tell the user, kindly and plainly, before confirming, and offer a separate post with a photo for Google. ⚠️ You CANNOT send a file the user attached to the chat : tool arguments are text, so a video or photo attachment cannot be transmitted. The user must give a public HTTPS link, or upload the file from the GMB Club …

delete_scheduled_post

ChatGPT
Use when the user wants to cancel or remove a planned post. Cancel one or several scheduled posts. TARGETS : pass post_ids[] to cancel many posts at once (ideal for "annule tous mes posts du week-end" → fetch list with list_scheduled_posts, then pass the ids). For a single post, pass post_id. Each id is checked individually : not-found / out-of-scope posts are skipped (reported), the rest are cancelled. Set confirm to true to actually cancel ; otherwise returns a preview.

edit_post_image

ChatGPT
Retouch an EXISTING image : takes image_url (the source) + prompt (the transformation). Identical to generate_post_image(source_image_url=..., prompt=...); prefer that one in new work. Kept so already-configured agents keep working. The source is never modified ; a NEW image_url is returned. ⚠️ AI editing REDRAWS the image : it will not reproduce a logo or text pixel-exactly. For an official visual where a partner's logo must be EXACT, generate the background here, then have the real PNG composited on top rather than asking the model to draw it.

generate_post_image

ChatGPT
Use when the user wants an AI-generated image for a post. Generate an image with AI from a text prompt. Returns image_url (an absolute HTTPS URL hosted on GMB Club). The URL is reusable as-is in any tool that accepts an image : (1) create_scheduled_post(image_url=..., image_mode='upload') for a social media post, OR (2) create_article(featured_image_url=...) for a WordPress/Wix article (set generate_image=false since the image is already provided), OR (3) update_scheduled_post(image_url=..., image_mode='upload') to attach to an existing pending post. Chain this tool with one of the above when the user wants 'a post/article WITH an image generated by AI' — don't say it's impossible, just generate then create. Counts against the user's image generation quota. BRAND ALIGNMENT (recommended): pass bulle_id to align the generated image with the establishment's visual identity. When provided, the tool loads the bulle's brand colors, preferred style, elements to include/exclude AND its business brief (identity + offer), then enriches the prompt automatically — same pipeline as Auto-Pilot. Without bulle_id, the prompt is used as-is (still works for one-shot creative images).

generate_report

ChatGPT
Use when the user wants to generate a performance report (PDF) for one or several listings. Trigger a new report generation. type controls what is generated : - 'fiche' : custom-period report for ONE Google Business listing (requires fiche_id, date_start, date_end YYYY-MM-DD). - 'combine' : multi-listing comparison report (requires fiche_ids[] with at least 2 entries, date_start, date_end). - 'journal' : send the email digest right now for one bulle (requires bulle_id, schedule='daily'|'weekly'|'monthly'). All types are async : the report (or email) is prepared in the background and shows up in list_reports / arrives by email shortly after. Without confirm=true, returns a preview of what would happen.

get_acquisition_stats

ChatGPT
Use when the user asks how their review request campaigns perform: requests sent, clicks, new reviews. Aggregate stats for review acquisition requests on a fiche : total sent, sent by channel (sms/email), sent by status, last 30 days trend.

get_article

ChatGPT
Use when the user asks to read or review one specific article. Get the current state of an article (read-only). Returns title, target_keyword, tone, language, status, featured_image_url, content_html (the full rendered HTML body), content_text (same body stripped of tags, ready to read or transform), excerpt, meta_title (SEO title used by Yoast/RankMath), meta_description, scheduled_at, published_at, word_count and destination. USE THIS as the canonical way to read article content before modifying it : if the user asks for a rewrite, tone switch, voice change, translation, or any other transformation, call get_article first to get content_html, transform it on your side, then send the updated HTML back via update_article(content_html=...). Never ask the user to paste their article manually. Also used by the article_preview widget to refresh itself once the async generation finishes. Status values : draft / queued / generating / ready / scheduled / published / error. When status=queued or generating, the content is still being written by AI in the background (1-3 min typical). When no_site_connected=true, the article is NOT visible online whatever its status : no website is linked to GMB Club. Say so to the user instead of calling it published. When summarizing to the user, refer to the article by its title, never by numeric ID.

get_bulle

ChatGPT
Use when the user asks for the details or settings of one named location. Get detailed info about a single bulle, including GMB fiches count and effective tone/language (inherited from sphere if not set).

get_gmb_metrics

ChatGPT
Primary tool for Google Business Profile performance. Use when the user asks: how is my listing doing, profile views, clicks, calls, direction requests. Fetch Google Business Profile performance metrics for a fiche : views (search/maps), searches (direct/discovery), and actions (calls/website/directions). Period in days (1-90, default 30). ALSO returns search_keywords: the real queries people typed to find this listing, aggregated over up to 18 months of stored history. Use them whenever the user asks what people search for, or asks for content based on how customers find them. A keyword flagged below_publication_threshold had fewer than 15 impressions that month and Google withheld the exact figure; never read it as zero. Requires a valid GMB OAuth connection on the fiche. fiche_id is the id returned by list_gmb_fiches.

get_gmb_optimizer

ChatGPT
Use when the user asks how complete or optimized their Google listing is, or what to improve on it. Get GBP optimization data for a Google Business Profile listing. Two modes via kind : kind='score' (default, fast) : read the latest cached score (0-100) with breakdown by category (description, categories, hours, attributes, photos, services, posts, reviews) directly from the database. No Google API call, instant. Returns null if no audit was ever run. kind='audit' (full, slow) : run a fresh full audit; fetches current state from Google, computes a 0-100 score with breakdown, generates concrete AI-powered suggestions for improvement (description, categories, hours, photos, attributes, services, posts). Cached for performance ; set force_refresh=true to bypass the cache when the user just applied changes and wants the new score. Pair with apply_gmb_optimizer_suggestions to push specific suggestions back to Google. Aliases : 'latest'/'current' = 'score' ; 'full'/'fresh'/'refresh' = 'audit'.

get_reviews_summary

ChatGPT
Use when the user asks for review statistics: how many reviews, average rating, rating distribution, response rate, trends. Aggregate stats over reviews : average rating, total count, rating distribution (1-5 stars), responded count and %. SCOPE : pass fiche_id to scope to one Google Business Profile listing, OR omit it to aggregate ACROSS EVERY fiche the user has access to. When cross-scoped, also returns per_fiche breakdown.

get_seo_data

ChatGPT
Primary tool for SEO results. Use when the user asks: my SEO rankings, keywords, competitors, site audit, PageSpeed scores. Read SEO data for a bulle. Three modes via kind (all read-only, no quota consumed) : kind='snapshot' (default) : the latest full SEO snapshot: positioned keywords, organic competitors, top pages, PageSpeed scores, AI analysis if available. Returns null if no scan was ever run (use run_seo(kind='scan') first). kind='competitors' : compact view of just the organic competitors (competitor domains, shared keywords count, traffic estimate) extracted from the latest snapshot. Use this when you only need competitor data without the full snapshot payload. kind='gsc' : Google Search Console metrics. Requires GSC OAuth connection on the bulle (configured in the dashboard). Pass gsc_kind to choose the view : 'queries' (top search queries, default), 'pages' (top landing pages), or 'traffic' (monthly clicks/impressions). The answer carries data_end (last day measured by Google) and is_current : when false, never present these figures as this month's. A non-null access_issue means Google withdrew access : the user must reconnect Search Console in the app. Aliases : 'latest'/'full' = 'snapshot' ; 'organic_competitors' = 'competitors' ; 'search_console'/'google_search_console' = 'gsc'.

get_social_stats

ChatGPT
Use when the user asks for social media performance numbers or a data-driven social strategy (what works best, engagement, follower growth). Reads REAL statistics for the connected social accounts (Facebook, Instagram, Pinterest, LinkedIn, TikTok): followers, per-post engagement (reactions/likes, comments, shares, reach, saves), account insights, and daily history series accumulated in DB. ALWAYS call this before recommending a social strategy and ground every advice in these numbers; never invent metrics. DATA HONESTY: a null metric means the connected account no longer grants permission to read it: report it as « indisponible », NEVER as zero, and suggest reconnecting the account from the Settings page (each connection carries data_quality notes explaining this). Platform limits: Instagram history is capped at 30 days by the API; LinkedIn stats require a company page; TikTok exposes profile totals only (no history); follower history builds day by day (no backfill). SCOPE: pass sphere_id and/or bulle_id to filter, platform to target one network. history_days: 7, 30 or 90.

get_visibility_scan

ChatGPT
Use when the user asks about their local ranking or visibility on Google Maps around their location, or about their visibility in AI answers (ChatGPT, Gemini, Google AI Mode, Perplexity, Claude). kind='maps' (default): the latest Google Local Pack visibility scan for a fiche (visibility_score, scanned_at). kind='ai': the latest AI visibility reading: per engine, how often the business is named in AI answers to real customer questions (citation rate with a confidence range, never a rank), every question asked, what each AI says about the business (brand_answers), every competitor the AIs name instead, every source they rely on, the history of comparable readings and the automatic monthly follow-up state (auto_follow). include_answers=true adds the full text of every AI answer of the reading (large: about 120 answers); question='<part of a question>' adds only the answers to that question. Both are kind='ai' only.

list_articles

ChatGPT
Use when the user asks about their blog articles: drafts, scheduled, published. List articles. SCOPE : pass site_id to scope to one WordPress site, OR pass wix_site_id to scope to one Wix site, OR omit BOTH to list ALL articles the user has access to (across every WordPress + Wix site of every sphere of the token + shared bulles). Optional filters : status (draft | queued | generated | ready | scheduled | published | error | failed), limit 1-100 (default 50). Returns id, title, target_keyword, status, word_count, created_at, scheduled_at (ISO 8601 UTC, when status='scheduled'), published_at, wordpress_site_id / wix_site_id, destination and no_site_connected. An article with no_site_connected=true is NOT visible online, whatever its status says : no website is linked to GMB Club. Say so to the user instead of calling it published. When summarizing to the user, list articles by title only; never expose numeric IDs (no '#123', no 'article 299'). Keep IDs strictly internal for follow-up tool calls.

list_bulles

ChatGPT
Primary tool to discover the user's business locations. Use when the user asks: my establishments, my locations, which businesses do I manage. List bulles (establishments). SCOPE : pass sphere_id to scope to one sphere (includes bulles shared into it), OR omit it to list EVERY bulle the user has access to (across every sphere of the token + shared bulles). Use the no-scope form when the user asks 'all my establishments'. is_shared=true + source_sphere_id mark bulles owned by another sphere.

list_gmb_fiches

ChatGPT
Use when the user asks about their Google Business Profile listings or whether they are connected. List Google Business Profile fiches (locations). SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ALL fiches the user has access to (across every sphere of the token, including bulles shared from other spheres). Essential when the user asks 'all my GMB listings' on multi-establishment accounts. bulle_id is an additional filter on the matched fiches. Returns id, name, address fields, sphere_id, bulle_id and connection status (has_token). id (accounts/…/locations/…) is the fiche_id every other tool expects; gmb_location_id is Google's bare number.

list_qr_codes

ChatGPT
Use when the user asks about their review QR codes and scan statistics. List QR codes. SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ALL QR codes the user has access to (across every sphere of the token + shared bulles). bulle_id is an additional filter. Returns name, destination URL, scan count, last scan date, and gate status.

list_reports

ChatGPT
Use when the user asks for their monthly performance reports. List monthly reports. SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ALL reports the user has access to (across every sphere of the token + shared bulles). type='fiche' returns per-fiche monthly reports ; type='combine' returns multi-fiche combined reports. Returns name, period, status and PDF/CSV URLs. status is 'ready', 'generating' (no file yet) or 'failed' (the generation did not complete: offer to generate the report again).

list_reviews

ChatGPT
Primary tool for reading Google reviews. Use whenever the user asks to see, list, read, filter or analyze reviews: latest reviews, negative reviews, unanswered reviews, what customers are saying. List reviews. SCOPE : pass bulle_id to scope to one location (bulle), fiche_id to scope to one Google Business Profile listing, OR omit both to get reviews ACROSS EVERY fiche the user has access to (all spheres of the token + shared bulles). Essential when the user asks 'all my 1-2★ reviews of the week' on a multi-fiche account. NEVER ask the user for an id : when they name a location, resolve bulle_id yourself via list_bulles (match by name), or simply call this tool without any id and filter. Optional filters : status='responded'|'unresponded', stars=1..5, limit (max 100, default 50). Returns review_id, customer_name, rating, text, responded, dates (+ fiche_id and fiche_name when cross-scoped). When summarizing to the user, refer to each review by author name, rating, fiche name, or the first words of the comment; never expose internal IDs (no '#abc123'). Keep IDs strictly internal for follow-up tool calls.

list_scheduled_posts

ChatGPT
Use when the user asks what posts are planned or already published: my posts, my publication calendar, what is scheduled this week. List scheduled social media posts. SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ALL scheduled posts the user has access to (across every sphere of the token + bulles shared from other spheres). Use the no-scope form when the user asks 'all my pending posts this week' on a multi-establishment account. Optional status filter (pending|queued|published|partial_published|partial_error|error|failed). Deleted posts are never returned. Limit 1-100, default 50. A published story carries story_stats per network: Instagram figures (views, reach, shares, interactions, follows, profile visits, navigation) are only given by Instagram during the story's 24 hours, then kept as final; too_few_viewers means under 5 views, not zero; replies are counted from the inbox; Facebook gives no story statistics (not_provided). When summarizing to the user, refer to each post by its content (first words, date/time, platforms) or by the bulle's name; never expose numeric IDs (no '#123', no 'post 540'). Keep IDs strictly internal for follow-up tool calls.

list_sites

ChatGPT
Use when the user asks which websites (PrestaShop, Shopify, Wix, Webflow, WordPress) are connected. List website integrations the user can publish articles to. Supported platforms : PrestaShop, Shopify, Wix, Webflow, WordPress. PLATFORM (optional filter) : pass platform='prestashop' / 'shopify' / 'wix' / 'webflow' / 'wordpress' to scope to a single platform, OR omit it to list ALL sites across every platform the user has access to. Aliases accepted : 'wp' = 'wordpress', 'ps' = 'prestashop'. SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ACROSS every sphere of the token + shared bulles. bulle_id is an additional filter that further restricts to one establishment. Returns a flat list where each item carries a platform field ('wordpress' | 'wix' | 'shopify' | 'prestashop') so you can route downstream calls (create_article uses site_id for WordPress, wix_site_id for Wix, etc.). Common fields : id, site_url, site_name, sphere_id, bulle_id, is_shared. Platform-specific fields : has_api_key / wp_version / plugin_version (WordPress, PrestaShop) ; has_token (Wix, Shopify) ; shop_domain (Shopify) ; module_version / is_connected (PrestaShop). EMPTY RESULT : when no site is connected at all, the list holds ONE item with error='no_site_connected', the website on file for each location, a connect_url and steps. NO SITE CONNECTED : when this tool returns error='no_site_connected', tell the user in plain words that no website is linked to GMB Club yet, so their articles cannot appear online, then walk them through steps. Never describe such an article as published or online. When a platform filter hides sites of other platforms, the item says error='no_site_for_platform' and lists other_platforms.

list_social_connections

ChatGPT
Use when the user asks which social networks are connected for a location. List active social media OAuth connections. SCOPE : pass sphere_id to scope to one sphere, OR omit it to list ALL connections the user has access to (across every sphere of the token + shared bulles). bulle_id is an additional filter. Returns platform, account_name, account_id, is_active, token_expires_at.

list_spheres

ChatGPT
Use when the user asks about their brands, workspaces or overall account structure. List the user's authorized spheres (brands/workspaces) for this connector. Returns id, name, slug, default tone and language.

list_tracked_keywords

ChatGPT
Use when the user asks which keywords are tracked and their current positions. List tracked SEO keywords. SCOPE : pass bulle_id to scope to one establishment, OR omit it to list ALL tracked keywords across every accessible establishment of the token. Each result includes bulle_id + bulle_name when cross-scoped. Returns id, keyword, current_position, monthly_search_volume, last_checked_at.

publish_article

ChatGPT
Use when the user wants an article published (or scheduled) on their website. Push a finished article to its publication target: WordPress site, Wix site, or direct API mode (when no external site is attached). The backend routes automatically based on the article's wix_site_id / wordpress_site_id. The article must have its content already written (use create_article with generate_content first if not). Set confirm to true to actually publish ; otherwise returns a preview with the full content text so the user can review before pushing it live. SCHEDULING : pass scheduled_at (ISO 8601, UTC or with timezone, ex: '2026-06-23T09:00:00Z' or '2026-06-23T11:00:00+02:00') to queue the article for automatic publication at that date instead of publishing now. The cron beat publish_scheduled_articles picks it up. scheduled_at must be in the future ; for now/past use just leave it unset. Confirm must still be true to actually queue. Tell the user the planned publication date in human form when summarizing the result. RETRY : an article whose previous publication failed (status failed) can be published again with this tool once the cause is fixed (site back online, content written via update_article…). REFUSAL WITHOUT A SITE : an article attached to no website, when no integration of the customer fetches GMB Club articles, is refused with error='no_site_connected' (or 'article_not_attached_to_site' when the location does have a connected site), in preview, confirmed and scheduled mode alike. NO SITE CONNECTED : when this tool returns error='no_site_connected', tell the user in plain words that no website is linked to GMB Club yet, so their articles cannot appear online, then walk them through steps. Never describe such an article as published or online.

publish_now

ChatGPT
Use when the user wants an already scheduled post to be published immediately. Publish a pending scheduled post immediately (skip the schedule). Without confirm=true, returns a preview of what would be published. Requires scope=write.

query_search_console

ChatGPT
Use when the user asks about real Google traffic: clicks, impressions, top queries, top pages (Search Console). Query Google Search Console LIVE for a bulle, reusing the stored OAuth credentials (no extra connection). Use for strategic questions the cached view can't answer : filter by query / page / country / device, custom date range, custom row limit. Required : bulle_id. Common patterns : - Top queries last 90 days : dimensions=['query'], days=90 - Top pages for a brand term : dimensions=['page'], filter_query_contains='nom-marque' - Daily traffic curve : dimensions=['date'], days=30 - Device split for a page : dimensions=['device'], filter_page='https://...' Optional params : - dimensions : list[str] parmi 'query', 'page', 'country', 'device', 'date' - days : nb de jours en arrière depuis aujourd'hui (défaut 28) - start_date / end_date : YYYY-MM-DD (override days si fournis) - row_limit : max 1000 - filter_query_contains / filter_query_equals : restreint sur la requête - filter_page_contains / filter_page_equals : restreint sur l'URL - filter_country : code pays ISO 3 lettres (FRA, USA…) - filter_device : 'desktop' | 'mobile' | 'tablet' - search_type : 'web' (défaut), 'discover', 'image', 'video', 'news' Renvoie : la liste brute des rows GSC avec clicks, impressions, ctr, position pour chaque combinaison de dimensions. Si la bulle n'a pas encore connecté Search Console, renvoie un message explicite.

reply_review

ChatGPT
Use when the user wants to publish a reply to a Google review. Post a reply to one or several Google reviews. TARGETS : pass replies[] for batch mode: a list of {review_id, text} dicts (each reply can have its own text, ideal for "réponds aux 20 avis 5★ d'hier avec ces brouillons"). For a single reply, pass review_id + text as direct params. Each review is checked individually : not-found / out-of-scope / tier-blocked reviews are skipped (reported), the rest are posted. Set confirm to true to actually post ; otherwise returns a preview of every reply for validation. MANY REVIEWS : for more than 10 reviews, or "all my reviews", do NOT write the replies yourself: pass all_unanswered=true with fiche_id (and limit if the user gave a number). The app writes each reply with the business's own instructions as it goes; the preview returns the counts, the duration, the reviews left to the owner, and examples. Above 5 replies nothing is posted on the spot: replies go out in the background, at most 25 per hour per listing, 8am to 10pm local time, because Google can reject replies posted in bursts.

request_media_upload

ChatGPT
Use when the user wants to publish a photo or a video THEY have, and you have no public HTTPS link for it, typically when they attached the file to this conversation, or say 'the video is on my phone'. You cannot read an attachment's bytes, so this tool gives the user a link to drop the file themselves. TWO STEPS : 1. Call with kind (no token) → returns upload_url. Give that link to the user and ask them to open it and drop their file(s) (it works on a phone too). The link is single-use and expires in 60 minutes. • kind='image' → one photo • kind='video' → one video • kind='carousel' → 2 to 10 images AT ONCE, for a carousel post. Use this instead of generating one link per slide. 2. When the user says it's done, call again with the token you got → returns status. Once status is 'uploaded' : • photo → create_scheduled_post(image_url=media_url, image_mode='upload') • video → create_scheduled_post(video_url=media_url) • carousel → create_scheduled_post(image_urls=media_urls): pass the WHOLE list, in order ; passing only media_url would publish a single slide. If status is still 'pending', the file has not arrived yet: tell the user, don't invent a URL. Prefer a direct public HTTPS link when the user already has one : upload_image(url=...) or create_scheduled_post(video_url=... / image_urls=[...]) are one step instead of three. And when YOU are the one producing the slides, call generate_post_image once per slide and skip this tool entirely. Requires scope=write.

run_seo

ChatGPT
Use when the user wants to launch a new SEO scan, site audit or keyword lookup. Trigger an SEO action for a bulle. Three modes via kind (all count against the SEO monthly quota; set confirm=true to actually consume, otherwise returns a preview). kind='scan' : full SEO snapshot (positioned keywords, organic competitors, top pages, PageSpeed scores). Result is then readable via get_seo_data(kind='snapshot'). Typical quota : 5 scans/bulle/month. kind='onpage' : multi-page OnPage audit (crawls up to max_pages pages, default 100, max 200). Returns a snapshot id to poll later. Use this to detect technical SEO issues (missing meta, broken links, duplicate content, etc.). kind='keyword_lookup' : on-demand single-keyword research; returns Google position, monthly search volume, difficulty. Requires the keyword param. Typical quota : 20 lookups/bulle/month. Aliases : 'full'/'snapshot' = 'scan' ; 'crawl'/'onpage_audit' = 'onpage' ; 'keyword'/'lookup'/'lookup_keyword' = 'keyword_lookup'. Requires scope=seo:run.

send_review_request

ChatGPT
Use when the user wants to ask a customer for a Google review by SMS, email or WhatsApp. Send a review request to ONE or SEVERAL customers (SMS or email). The customer(s) receive a personalized link to leave a Google review. TARGETS : pass contacts[] for batch mode: a list of dicts, each containing customer_name + channel ('sms'|'email') + customer_phone OR customer_email + optional custom_message. Example contacts entry: {'customer_name': 'Marie', 'channel': 'sms', 'customer_phone': '0612345678', 'custom_message': '...'}. Ideal for "envoie une demande à ces 50 clients de la liste". All contacts in the batch target the same fiche_id. For a single send, pass fiche_id + customer_name + channel + customer_phone/email directly. custom_message lets you override the default text. Inside the message, four variables are auto-replaced: {nom_client}, {nom_etablissement}, {lien_avis}. Example: 'Bonjour {nom_client}, merci pour votre visite chez {nom_etablissement}. Donnez votre avis ici : {lien_avis}'. If {lien_avis} is omitted, the email still shows the review button. Without confirm=true, returns a preview of every contact for validation. Requires scope=acquisition:send.

suggest_review_reply

ChatGPT
Use when the user wants a draft reply to a review WITHOUT publishing it. Generate an AI-suggested reply for a single Google review, in the review's OWN language (auto-detected). Uses the bulle's tone, signature, business name and length settings, exactly like the « Suggestion IA » button of the GMB Club dashboard. Read-only on the review itself (does NOT publish; to publish, pass the returned text to reply_review with confirm=true). Returns suggested_response, customer_name, rating, business_name. Primarily called by the reviews_list widget when the user clicks « Rédiger une réponse »; gives a draft in the right language (EN review → EN draft, FR review → FR draft, etc.). When summarizing to the user, refer to the review by author name or rating, never by numeric ID.

update_article

ChatGPT
Use when the user wants to edit an existing article: title, content, image, date. Update fields of an existing article: title, content_html, excerpt, meta_title (SEO title, 60 chars recommended), meta_description (155 chars recommended), target_keyword, featured_image_url. Pass only the fields you want to change ; the others stay as they are. meta_title and meta_description are pushed to WordPress via Yoast/Rank Math (and to Wix via equivalent fields), so editing them here updates the live SEO tags after the next publish. TO REWRITE OR TRANSFORM EXISTING CONTENT (e.g. switch tone, translate, add a section, change voice from 'vous' to 'tu') : FIRST call get_article(article_id); it returns the full content_html AND the stripped content_text of the article. Take that content_html as your source, apply your transformation on your side, then send the full updated HTML back via content_html. Never tell the user the content cannot be retrieved and never ask them to paste it manually; get_article is always available for this. Never call publish_article just to read content ; publish_article is for pushing live, get_article is for reading. Cannot edit articles already published live (status='published'). Set confirm to true to apply ; otherwise returns a preview of the changes. When summarizing results to the user, refer to the article by its title, never by numeric ID.

update_scheduled_post

ChatGPT
Use when the user wants to edit a planned post: change text, date, image, or the Google event / offer (its hours, title, button, coupon). Update a scheduled post or a draft. Only editable while it has not been published yet. Fields : content, scheduled_at, image_url, image_mode, and the gmb_* options below (only the ones you pass are changed). STATUS : status='draft' keeps the post from being published (it stays in the calendar), status='scheduled' sends a draft to publication at its date (which must be in the future). TIMEZONE : scheduled_at follows the same rule as create_scheduled_post: no tz info → interpreted as local time of the post's bulle/sphere ; with 'Z' or offset → respected as-is. IMAGES : image_url can be the URL returned by generate_post_image or upload_image (base64 or HTTPS url). CAROUSEL : pass image_urls=[url1, url2, ...] (2 to 10 images) to replace the post's media with a multi-image carousel (same rules as create_scheduled_post). VIDEO : pass video_url (public HTTPS link, MP4/MOV/AVI/WebM, max 100 MB) to replace the media with a video. FORMATS : formats={'instagram': 'story'} turns a platform into a story (or back to 'post'), same rules as create_scheduled_post ; refused on a platform already published. GOOGLE BUSINESS POST TYPE (only used when 'gmb' is a platform) : gmb_topic_type = STANDARD (news, default) | EVENT | OFFER. EVENT and OFFER need gmb_event_title and gmb_event_start_at (gmb_event_end_at optional, same day if omitted). gmb_event_start_at / gmb_event_end_at are the event's hours AS SHOWN IN THE STORE, local wall-clock time, no timezone conversion (e.g. '2026-10-08T10:00'). They are NOT the publication date (that is scheduled_at). Each establishment can have its own hours : update each post separately. Button : gmb_cta_type = BOOK | ORDER | SHOP | LEARN_MORE | SIGN_UP | CALL with gmb_cta_url (not for CALL, not for OFFER). Offer details : gmb_offer_coupon_code, gmb_offer_redeem_url, gmb_offer_terms. Pass an empty string to remove a field. YOUTUBE / TIKTOK OPTIONS : platform_options={'youtube': {...}, 'tiktok': {...}}, only read for platforms in the post. youtube: privacy_status (public | unlisted | private): Who can watch the video. REQUIRED when YouTube is targeted and the post is scheduled: nothing is ever published private by default. is_short (true/false): true adds #Shorts to the description. YouTube classifies a Short by its format (vertical or square, 3 minutes max), the hashtag is only a hint. title (text, max 100): Video title (max 100 chars). Default: first line of the YouTube caption. description (text, max 5000): Video description (max 5000 chars). Default: the YouTube caption. tags (list): Keyword tags (list of strings; YouTube keeps 500 characters in all). made_for_kids (true/false): COPPA declaration: true if the video is made for children. Default false. category_id (text, max 10): Numeric YouTube category id (e.g. '22' = People & Blogs). Optional. tiktok: privacy_level (PUBLIC_TO_EVERYONE | MUTUAL_FOLLOW_FRIENDS | FOLLOWER_OF_CREATOR | SELF_ONLY): Who can see the post. REQUIRED when TikTok is targeted and the post is scheduled (TikTok forbids a default). Must be a level the TikTok account allows: a private account cannot publish PUBLIC_TO_EVERYONE. title (text, max 2200): Caption shown on the video, hashtags included (max 2200 chars; 90 for a photo post). Default: the TikTok caption. disable_comment (true/false): true = comments off. Default true (TikTok rule: interactions are off unless the user allows them): pass false to allow comments. disable_duet (true/false): true = Duets off. Default true: pass false to allow Duets (video only). disable_stitch (true/false): true = Stitches off. Default true: pass false to allow Stitches (video only). auto_add_music (true/false): true lets TikTok add a trending sound. Default false. branded_content (true/false): Paid partnership promoting a third-party brand ('Paid partnership' label). Never with SELF_ONLY. brand_organic (true/false): Promotes the creator's own business ('Promotional content' label). description (text, max 4000): Ph…

upload_image

ChatGPT
Use when the user provides an image file/URL to attach to a post. Bring an image into the publication pipeline (returns image_url usable in create_scheduled_post with image_mode='upload'). Accepts EITHER image_base64 OR url (the user already has the image hosted somewhere publicly). Exactly one of the two. ⚠️ AN IMAGE THE USER ATTACHED TO THE CHAT CANNOT BE FORWARDED. You see attachments as rendered images, not as raw bytes you can transcribe, and tool arguments are text, so there is no way to fill image_base64 from an attachment. Do NOT guess, invent or approximate base64 content. Tell the user plainly: they must give a public HTTPS link to the image, or upload the file from the GMB Club dashboard. image_base64 only works for image data you already hold as text (e.g. produced programmatically). ONE EXCEPTION: an attachment address issued by the GMB Club assistant itself (…/static/uploads/… on the app's own domain) is readable here: pass it as url. PREFER url in every other case : no base64 round-trip through the chat transport, much faster, never stalls on large payloads. When using url : HTTPS only ; redirects, private/loopback/link-local IPs and non-image content types are refused (anti-SSRF). Accepts JPEG, PNG, WebP, GIF up to 8 MB raw. Use generate_post_image instead when the user wants the AI to create the image from a text prompt. Requires scope=write. Pass bulle_id when you already know which location the image is for: if that location has 'brand logo on visuals' enabled, the logo is stamped onto the image. Omit it and the logo is only applied when the account has a single location. DESTINATION. destination='post' (default) : the image enters the publication pipeline, as above. destination='mediatheque' : the photo is KEPT in the location's media library (the real photos of the business: storefront, products, team, premises, work done) that the AI reuses as visual references for future visuals and as knowledge about the business. Use it when the user shares a photo of their business to keep, or asks to add a photo to their media library. Requires bulle_id. Pass description when the user says what the photo shows (otherwise the AI describes it). Returns media_item_id ; the photo is never attached to a post by itself.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/gmb-club"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/gmb-club)