Almost everything written about Meta ads and Claude covers the same half of the job: plug your own ad account in, read your own metrics, pause your own campaigns. Useful, but it answers "how is my stuff doing?" and never "what should I run next?" When you connect Meta Ads to Claude properly, you wire up both directions: your account data flowing in, and the entire public Meta Ad Library, every competitor's active ads, flowing in beside it.
This guide is the setup page for the second half, with an honest map of the first. If you already have the connection and want the full research workflow inside Claude, the companion piece is the Claude + Meta ads tutorial. This post gets you connected and pointed at the right first prompts.
What "connect Meta Ads to Claude" actually means
There are two different connections hiding under one phrase, and most people only ever hear about the first.
Direction one: your account data in. An MCP server with access to your Meta ad account lets Claude read campaign performance, shift budgets, and pause or resume things conversationally. Meta ships an official MCP for exactly this at mcp.facebook.com/ads. If your only goal is managing an account you already run, that is the right tool and we say so plainly further down.
Direction two: competitor intelligence in. The most valuable Meta ads data is not in your account. It is in the Meta Ad Library, where every active ad from every brand sits in public view, and where the browsing experience is so bad that "public" barely counts. An MCP that ingests the Ad Library gives Claude something no account connection can: evidence of what is already working for everyone else in your category. That is the side AdWhispr was built for, and we made the full case for the research half here.
The honest framing: direction one tells you how your ads are doing, direction two tells you what to run next. You want both. Here is the setup for the research side, which takes about two minutes.
Connect AdWhispr to Claude in two minutes
No code, no tokens to copy. The connection is a standard custom connector with OAuth.
Step 1: Open the connector settings in Claude. In Claude, go to Settings, then Connectors, then Add, then Add custom connector. Connectors are available on Claude.ai in the browser and in the desktop app.
Step 2: Paste the AdWhispr MCP URL.
https://adwhispr.com/api/mcp
Name it AdWhispr so it is easy to spot in the tool list later.
Step 3: Authorize. Click Connect. Claude opens an AdWhispr sign-in window. Approve it and the connector goes green. This is a normal OAuth handshake: AdWhispr never sees your Claude conversation, and Claude never sees your AdWhispr password. A free AdWhispr account is enough, no credit card.
Step 4: Confirm it works. Start a new chat and paste:
Use AdWhispr to research Gymshark's Meta ads and show me their longest-running winner.
Claude should call the tools, pull the live ad library, and come back with the oldest ad still running plus how many days it has been live. If Claude answers from general knowledge instead, start a fresh chat and name AdWhispr explicitly.
Using Claude Code instead? One command replaces the settings screen:
claude mcp add --transport http adwhispr https://adwhispr.com/api/mcp
Using Claude Desktop with a local config? The npm package writes it for you:
npx adwhispr-mcp-server config
Restart the app and the same tools show up. Every path connects to the same server.
Your first fifteen minutes: one competitor, start to clone
Here is the loop we would run on day one. It goes from a Facebook page URL to a launch-ready creative without leaving the chat.
1. Ingest a competitor from a single Facebook page URL. Paste the URL or just name the brand. AdWhispr resolves the page, ingests their entire Meta Ad Library presence, every active ad, not a sampled handful, and snapshots it daily from then on.
2. Ask for the longest-running ads. Meta's Ad Library does not publish CTR, CPC, or ROAS, so AdWhispr never prints them. What the library does reveal is how long an ad has been running, tracked via those daily snapshots. Brands do not keep paying for losing ads, so days-running is the honest performance proxy. An ad live for 150 days has survived months of someone else's optimization budget.
3. Extract the hooks. Every ingested ad gets AI classification: hook, format, tone, offer. Instead of squinting at thumbnails, you ask what hooks the brand leans on and get an answer built from their whole library.
4. Clone the winner. clone_ad takes a proven image ad and produces a version adapted for your brand; clone_video_ad does the same for video. You are iterating on a market-validated concept instead of briefing from a blank page. The full research-to-live workflow is in clone a competitor's ad and launch it.
The same ingestion covers Instagram placements too, since Meta's Ad Library spans both. If Instagram is where your category fights, start with finding competitors' Instagram ads.
Six prompts to paste in
Add [competitor's Facebook page URL] and show me their ten longest-running active ads.
What hooks has [competitor] leaned on this quarter, and which formats carry them?
Compare [brand A] and [brand B] on Meta. Who is iterating faster, and whose offer changed recently?
Show me the longest-running UGC ads from [category] brands, with the hook on each.
Take [competitor]'s longest-running ad and clone it for my brand.
Generate a brief on [competitor]'s last 90 days of Meta ads: hook taxonomy, format mix, days-running distribution, verdict.
Research prompts are read-only and touch no ad account, so they are the safe place to start.
The honest landscape: what each connection can and cannot do
Meta's official Ads MCP (mcp.facebook.com/ads). This is Meta's own server for managing your own ad account from an AI assistant. For pure account management, reading your performance, adjusting your campaigns, it is the natural choice: first-party, maintained by Meta, no middleman. Its gap is structural, not a bug: it has zero competitor research. It cannot see the Ad Library of your rivals, so it can never tell you that a competitor has run the same hook for 200 days or that three brands in your category just switched to the same offer.
Reporting and analytics connectors. Various MCP servers and integrations pipe your Meta performance data into dashboards or spreadsheets. Good for stakeholder reporting; same blind spot, they only see accounts you own.
AdWhispr. Full Ad Library ingestion per brand, daily snapshots for days-running, AI classification and hook extraction, and cloning via clone_ad and clone_video_ad. On execution, plain honesty: launching Meta campaigns from AdWhispr is currently in beta. What is generally available on the launch side is Google (Search and Performance Max) and TikTok. So the combined play most of our users run: manage your Meta account with Meta's official tools or Ads Manager, use AdWhispr to know what to run based on competitor evidence, clone the winner, and launch the Google and TikTok side from the same chat. The Google half of that setup is covered in connecting Google Ads to Claude.
None of these tools replaces the others. The account connection answers "how is it going?" The research connection answers "what should I run?" The strategy edge lives in the second question, because it is the only input your competitors cannot see you using.
A note on safety
Connecting AdWhispr to Claude does not touch any ad account on its own. Competitor research reads public ad-library data. Anything that would spend money, connecting your own Google, TikTok, or Meta account and launching a campaign, is a separate, explicit step with its own OAuth approval. You cannot accidentally launch an ad by asking a research question.
FAQ
Is there an official Meta Ads MCP?
Yes. Meta ships one at mcp.facebook.com/ads, and it is the right tool for managing your own ad account from Claude: your campaigns, your budgets, your metrics. It has no competitor research at all. It cannot read the Ad Library of other brands, which is exactly the gap AdWhispr covers.
Can Claude see competitors' Facebook ads?
Yes, once a research connector is attached. AdWhispr ingests a competitor's entire Meta Ad Library presence from their Facebook page, snapshots it daily, and makes it queryable in plain English: longest-running ads, hooks, formats, offers, and cross-brand comparisons.
Can I launch Meta campaigns directly from Claude with AdWhispr?
Meta execution from AdWhispr is in beta, so treat Ads Manager or Meta's official MCP as the launch path for Meta today. Google Search, Performance Max, and TikTok launches are generally available from the same chat, which is where most users run the clone-and-launch leg.
Does this cost anything to try?
No. The free AdWhispr plan needs no credit card and includes enough monthly agent calls to connect, ingest a competitor, and run a real research pass. Paid plans lift the call limits and add clone credits.
Connect once, then ask better questions
Every media buyer already talks to Claude. The difference after this setup is what Claude can see: not just your account, but the public evidence of every ad your competitors are paying to keep alive. Connect the research side in two minutes, point it at your biggest rival, and ask what has been working for them the longest. That answer is the strategy input nothing inside your own account can give you.
Start free at adwhispr.com, then run your first prompt from the list above.
Your account tells you how it went. The Ad Library tells you what to run next.