A video clone waits at approve_video_clone before a credit is used, and campaigns are created paused with an explicit confirm before any budget moves — which matters when one conversation is doing the research, the creative and the launch. The research side is honest in a way that is easy to miss: what comes back is days-running plus hook and format distributions, signals anyone can check, rather than a ROAS figure nobody outside the advertiser could know. Two practical notes: add_brand resolves against the public ad library by the exact brand name and then ingests in the background for around 40 seconds before its ads are queryable, and Meta launching is still beta where Google and TikTok are not.
A bridge to AdWhispr's hosted server covering the whole loop: research a brand's Meta and TikTok ad libraries for the creative that has been running longest, clone it as an image or a video for your brand, then launch it on Google Search or Performance Max, TikTok, or Meta. The local package is a thin open-source wrapper — the data and the account live on AdWhispr and this stores nothing.
- Research: find_competitors returns brands verified to be advertising right now, ranked by live ad count; search_brands finds brands already tracked and add_brand starts ingesting a new one
- get_brand_ads returns ads with hooks, formats, days-running and creative links; get_brand_stats aggregates counts, hook and format distributions and spend ranges; search_ads is semantic search by concept such as "before/after" or "social proof"
- compare_brands puts two or three side by side, and generate_brief produces a competitive brief with the longevity curve, top hooks, format mix and a verdict
- research_tiktok_ads covers TikTok video, thumbnails and run dates; research_keywords returns Google volume, competition and CPC ranges, and research_competitor_keywords the terms a competitor's domain ranks and bids on
- get_my_brand, save_my_brand and clear_my_brand keep your own brand in context, while add_product, list_my_products and remove_product hold the catalogue behind per-product ads, with Shopify import
- Creative: clone_ad and clone_tiktok_ad clone a winner as a real generated image or video, clone_video_ad selects a model tier from cheap drafts to premium finals, clone_from_script and clone_from_concept start from your own words, and revise_video_brief edits the shot list before anything renders
- approve_video_clone is the confirm step — nothing renders and no credits burn until it runs — with check_video_status polling the render and list_my_creatives holding the results
- Launch: connect_ad_account runs the OAuth for Google, TikTok or Meta, list_ad_accounts and list_campaigns show what exists, launch_search_campaign and launch_pmax_campaign cover Google, launch_tiktok_campaign covers TikTok, and launch_campaign and launch_cloned_ad take a clone straight to air
- get_account_performance reads real numbers from the connected account, and update_budget, pause_campaign and resume_campaign manage what is live
- Campaigns are created paused by default with an explicit confirm before anything spends, and competitor research reports only verifiable signals — days-running rather than an invented ROAS
Npx adwhispr-mcp-server config writes the Claude Desktop entry; otherwise point any stdio client at npx -y adwhispr-mcp-server serve, or connect straight to https://adwhispr.com/api/mcp if your client speaks remote MCP with OAuth. Node 18 or newer. Sign-in opens a browser on the first tool call, and clearing ~/.mcp-auth resolves a stuck login loop. Free gives 15 agent calls a month with unlimited tracked brands and 1 clone, no card; Pro is $39/mo, or $31/mo billed annually, for 150 calls, 10 image clones and 10 video credits with launching on Google, TikTok and Meta; Agency is $149/mo, or $119/mo annually, for 3,000 calls, 50 image clones and 40 video credits with team seats. Google and TikTok execution are generally available, Meta is beta. ADWHISPR_MCP_URL points at a different server. MIT.
One command — npx adwhispr-mcp-server config