MCP & Automation

Facebook Ads MCP: Connect Claude and ChatGPT to Meta Ads

The four main ways to connect Claude or ChatGPT to Meta Ads through MCP, compared on writes, safety, platforms and price, with setup steps and prompts.

Facebook Ads MCP: Connect Claude and ChatGPT to Meta Ads
Definition Facebook Ads MCP
A Facebook Ads MCP (also called a Meta Ads MCP) is a Model Context Protocol server that exposes the Meta Marketing API as tools an AI assistant can call. Once it is connected, Claude or ChatGPT can pull campaign data, explain performance and, if the server allows writes, create or edit campaigns, ad sets and ads.

Facebook Ads MCP options compared

These are the four options most people searching for a Facebook or Meta Ads MCP end up choosing between. Verified on 18 Nov 2026 from each vendor’s pages.

Verified on 18 Nov 2026

Meta Ads MCP servers at a glance

 Meta official Ads MCPPipeboardGoMarbleWhathead
HostingHosted by Meta at mcp.facebook.com/adsHosted; source-available repo (BSL 1.1)Hosted; its open-source server is read-onlyHosted remote server at whathead.com/mcp
Ad platformsMeta onlyMeta, Google Ads, TikTok, LinkedIn and moreMeta, Google Ads and moreMeta plus 11 ad platforms, GA4, Search Console, Shopify
Create campaigns, ad sets, adsYes, created pausedYes, created pausedOn the Operator plan and upYes, previewed and confirmed before publishing
Safety modelPaused creates; agent rules in limited availabilityPaused creates; scoped tokensApproval gates and audit log (Operator and up)Dry run by default; confirmation for live and destructive calls; per-connector restrictions; call log
Cross-platform transferNot offeredNot listedNot listedYes: Meta to TikTok, Snapchat, X and others
PriceNo price listed during the betaFree (30 runs a week); Pro $29.90/monthCopilot $139/month (read-only); Operator $399/monthMCP Actions $39/month; 14-day trial
Best forMeta-only advertisersDevelopers and small teams on a budgetTeams that want an AI analyst plus writesTeams and agencies on several ad platforms

Which Facebook Ads MCP should you choose?

Meta’s official Ads MCP is the default if Meta is your only ad platform and a beta is acceptable. It is first-party, has about 88 tools across ad creation, reporting, catalogs, signals and tests, and creates new campaigns, ad sets and ads paused, so nothing spends until you activate it. It cannot delete campaigns, ad sets or ads, and it does nothing for Google, TikTok or Snapchat.

Pipeboard suits people who want a cheap hosted server and are comfortable wiring it up themselves. The free tier allows 30 tool executions a week; Pro is $29.90 a month with unlimited executions and three ad-account slots.

GoMarble is closer to an AI analyst than a plain server: diagnostics, reports and creative analysis, with MCP access on every plan. Writes such as launching campaigns or changing budgets start on the $399 Operator plan; the $139 Copilot plan is read-only.

Whathead fits when Meta is one of several platforms you run. One connection covers Meta, Google Ads, TikTok, Snapchat, LinkedIn, Reddit, X, Microsoft Ads, DV360, Criteo, Apple Ads and Amazon Ads, and the same workspace handles bulk edits, cross-platform transfers and Google Sheets reporting. It has no free plan, and if you only ever touch Meta, Meta’s own server costs nothing. More detail: Meta Ads MCP on whathead.com and Whathead vs Meta’s official MCP.

You can also run two. Some teams keep Meta’s server for Meta-only work such as catalogs and signals, and use a multi-platform server for cross-channel reporting and changes.

How to connect Claude to Meta Ads

The steps below use Whathead. Meta’s official server is added the same way with its own URL (https://mcp.facebook.com/ads) and a Facebook Login for Business sign-in instead of a Whathead account.

  1. Connect Meta in Whathead. Sign in at whathead.com on a plan with MCP access (the 14-day trial counts), open Integrations, connect Facebook with the profile that can access your ad accounts, and pick the ad accounts the workspace may use.
  2. In Claude, open Customize > Connectors, click + Add, then Add custom connector.
  3. Name it Whathead and enter the server URL https://www.whathead.com/mcp, then click Continue.
  4. Keep the recommended sign-in settings, click Add, then sign in with your Whathead account and approve the access popup.
  5. In a chat, click + in the message box, choose Connectors and switch Whathead on.
  6. Start read-only. Ask something that only reads data, for example “List my Meta ad accounts and yesterday’s spend by campaign.” Claude asks for permission the first time it calls each tool.

On Claude Team and Enterprise plans an owner adds the connector once under Organization settings > Connectors, and each member then clicks Connect. Menu names change from time to time; the current steps are in the Whathead connection guide.

How to connect ChatGPT to Meta Ads

  1. Connect Meta in Whathead as above.
  2. In ChatGPT on the web, open Plugins (chatgpt.com/plugins), select +, then Add custom MCP server.
  3. Enter a name, paste https://www.whathead.com/mcp as the server URL and choose OAuth. Accept the risk warning and create the plugin. In a workspace, an admin may need to allow custom servers first.
  4. Sign in with your Whathead account and approve the popup.
  5. In a conversation, type @ and pick Whathead. ChatGPT asks you to confirm any tool that is not marked read-only; Whathead marks its read tools as read-only, so reports run without extra prompts and writes always ask.

What the AI can do in Meta Ads

With a write-capable server connected, these jobs work well from a chat:

  • Reporting: spend, results, CPA and ROAS by campaign, ad set or ad, with breakdowns by age, placement or country and comparisons between date ranges.
  • Audits: ad sets with no conversion event, ads with missing UTMs, campaigns past their end date, names that break your convention.
  • Bulk changes: pause or resume a list of ad sets, change budgets by a percentage, extend end dates.
  • Builds: draft a campaign with ad sets and ads from a brief or a media plan, using creatives already in your library.
  • Cross-platform work (Whathead): copy a working Meta campaign to TikTok or Snapchat as a draft, or compare Meta with Google Ads in one answer.

What it can’t do, or shouldn’t

  • It only sees what the API exposes. Some Ads Manager features reach the Marketing API late or never, and every server is limited to what the API returns.
  • It can’t bypass Meta’s rules. Ad review, special ad categories, spending limits and account roles still apply.
  • It shouldn’t act on its own. A model can misread a metric or pick the wrong ad set when names look alike. Ask for the list of entities and the exact change before anything is written.
  • Meta’s official server can’t delete campaigns, ad sets or ads, and it only covers Meta.
  • Numbers are Meta’s view. Answers use Meta-reported conversions, which usually differ from GA4 or your MMP.

Safety checklist before you allow writes

Before the first live change

Keep a person in the loop

  • Run read-only prompts for the first week.
  • Connect only the ad accounts the assistant needs.
  • Keep new campaigns paused until a person has reviewed them.
  • Ask for a preview (dry run) of every change and read it before confirming.
  • Don’t click “Allow always” on write tools.
  • Use Meta’s agent rules or Whathead’s connector restrictions to cap what the assistant can touch.
  • Read the change log after each session.

In Whathead, write tools run as a dry run by default and return the exact change; live and destructive calls need an explicit confirmation, and every call is logged on the MCP page of your workspace. In Meta’s official server, new entities are created paused and activated separately, and admin rules that block certain agent actions are in limited availability.

Prompts to start with

  • “Which Meta campaigns spent more than $500 in the last 7 days with a CPA above $40? Show spend, results and CPA.”
  • “List ad sets that end in the next 3 days and suggest new end dates. Don’t change anything yet.”
  • “Find active ads whose URL has no utm_campaign parameter.”
  • “Prepare a 15% budget increase for the three ad sets with the lowest CPA this week, show me the preview and wait for my confirmation.”
  • “Copy campaign Spring Sale to TikTok as a draft and list the fields that need my input.”

Facebook Ads MCP: common questions

Is there an official Facebook Ads MCP?

Yes. Meta runs the Ads MCP server at mcp.facebook.com/ads. It entered open beta on 29 April 2026 and was opened to any Meta developer app on 16 July 2026. Meta lists no price for it during the beta.

Can Claude create Facebook ads?

Yes, through a write-capable MCP server. Meta’s server, Pipeboard and Whathead can create campaigns, ad sets and ads. Meta and Pipeboard create them paused; Whathead shows a preview and publishes only after you confirm.

Is it safe to give an AI write access to my ad account?

It is as safe as the guardrails you keep: only the accounts it needs, paused or previewed writes, explicit confirmation and a change log. Don’t let the assistant approve its own changes.

Do I need a Meta developer app?

Not for hosted servers such as Meta’s, Pipeboard or Whathead; you sign in with OAuth. Self-hosting an open-source server usually means creating your own Meta app and access token.

Does it work with ChatGPT or only Claude?

Both. Any client that supports remote MCP servers with OAuth works, including Claude, ChatGPT, Cursor and n8n.

How much does a Facebook Ads MCP cost?

Meta lists no price for its server during the beta. Pipeboard has a free tier and Pro at $29.90 a month. GoMarble starts at $139 a month (read-only) and $399 for writes. Whathead’s MCP Actions plan is $39 a month with a 14-day trial.

Sources

Written by Ahmed Shahin, founder of Whathead. Last reviewed 18 November 2026.