Last updated: July 23, 2026

A Meta ads MCP is a Model Context Protocol server that connects Claude to your Meta ad account, so you can pull performance reports, check creative fatigue, and brief new ad variants from a chat window. The fastest working setup in 2026 is a hosted server: you add a connector in Claude, authorize it, and run your first performance query a few minutes later, with no Marketing API application and no code.

That second sentence matters, because the DIY route is genuinely painful. Meta's Marketing API is powerful, but between app review, access tokens, and rate limits, most marketers who start building their own integration stop halfway. This guide covers what a Meta ads MCP actually does, the four ways to set one up, the prompts worth running every week, and the limitations nobody puts on a landing page. The Google side of the same workflow lives in the sister guide, Google Ads MCP.

What is a Meta ads MCP?

MCP stands for Model Context Protocol, the open standard Anthropic published for connecting AI assistants to external tools and data. A Meta ads MCP, and you will also see "Facebook ads MCP" for the same thing, is any MCP server that exposes Meta advertising capabilities as tools Claude can call: reading campaign and creative performance, auditing spend, and in the more complete implementations, generating and publishing new ads.

Without one, Claude knows a lot about Meta ads in general and nothing about yours. It can explain what a healthy CTR looks like, but it cannot see that your top campaign's frequency crossed 4 last Tuesday. An ads MCP server closes that gap. Claude reads your live account data, reasons over it, and, where the server supports write actions, acts on it.

For Meta specifically, the useful servers split into two camps. Data-out servers surface reporting and audits. Full-loop servers add creation: Claude briefs an agent that produces the ads, publishes them, and reads performance back. That second camp is where a Claude ads integration stops being a dashboard replacement and starts being an AI marketing agent you steer from chat.

What can Claude do once it is connected to Meta ads?

Taking the Superscale AI MCP, live on claude.ai since late June 2026, as the reference implementation, three workflows cover most of what teams actually run.

1. Context in, brand-tailored ads out. Claude pulls your brand guidelines, briefs, and reference assets from wherever they live, Notion, Google Drive, a spreadsheet, and hands them to the Superscale agent for a run. What comes back is ad copy, static images, and UGC video built on the brand and product context stored in Superscale, not generic output.

2. Ad data out. Because Superscale connects to Meta, TikTok, Google Ads, Shopify, and Google Analytics, the MCP surfaces live performance data into Claude: performance reports, competitive research, and scale, cut, and budget recommendations. This is everything after ad creation, media buying included.

3. Scheduled automation. Any run can be put on a cadence from Claude: a weekly creative refresh, a Monday variation batch, a Friday performance summary posted to Slack.

Under the hood, the tool surface looks like this:

Tool group What Claude can do with it
Orgs and brands List organizations and brands, pull the stored brand and product context
Agent runs Create a chat, send the message that triggers the agent, list generations
Outputs Show chat outputs, list assets, pull single or listed reports
Context in Upload reference media to attach images to a run
Ops Check parallel-run slot usage and the current user

You never need to name a tool. You describe what you want in plain language and Claude picks the calls.

The 4 ways to connect Claude to Meta ads

Route Setup time Technical level What you get Best for
DIY: Meta Marketing API + custom MCP server Days High, developer required Exactly what you build; you own tokens, scopes, rate limits Engineering teams with unusual requirements
Open-source Meta ads MCP servers 1 to 2 hours Medium, config files and an access token Mostly read access to account data inside Claude Technical marketers who like tinkering
Hosted ad-data MCPs (Ryze AI and similar) Minutes Low Reporting and audits through a remote connector URL Fast read-only analysis
Superscale AI MCP Minutes Low Creation (copy, statics, UGC video) plus performance data plus scheduling, published to Meta Teams that want the full loop in one connector

Choosing between them is simpler than the table suggests. If your only goal is asking questions of your spend data, a data-only route is enough, and the open-source options cost nothing beyond your evening. The DIY route makes sense when you have a developer and a requirement nobody else covers; Meta's Marketing API documentation is the starting point, and budget real time for app review and token management. The full-loop route earns its price when your bottleneck is creative production volume rather than analysis, which on Meta in 2026 it usually is. More on that under Andromeda below, and the wider tool landscape is ranked in the best AI media buying tools for Meta.

How to set up the Superscale AI MCP for Meta ads

Six steps, and only one involves a plan decision.

  1. Create a Superscale account and set up your brand. Brand, organization, and product setup stays inside Superscale; the MCP is the remote control, not the hub. The free plan includes 1,000 credits with no card required.
  2. Connect your Meta ad account inside Superscale. This is the plan gate: ad-account integrations for Meta, TikTok, and Google require the Advanced plan at $99/month or higher and are not available on Starter at $49/month. Creation works on any plan; connected-account reporting does not. Full tiers are on the pricing page.
  3. Add the Superscale connector in Claude. Add the Superscale MCP server in claude.ai's connector settings and authorize it. No JSON files, no local install.
  4. Connect your knowledge tools to Claude (optional). Notion and Google Drive connections let Claude pull briefs and brand docs and pass them into a run as context.
  5. Run your first prompt. Start with the performance report below, and confirm Claude sees the right organization and brand before anything else.
  6. Put the winners on a schedule. Once a prompt earns its place, ask Claude to run it on a cadence: weekly refresh, Monday batch, Friday summary to Slack.

Meta ads prompts to run in Claude

Three prompts cover a full weekly loop. Copy, paste, adjust the numbers to your account.

The weekly performance report

Pull performance for my Meta ad account for the last 14 days.
Break results down by campaign and by individual creative.
Flag anything with a CPA more than 20 percent above the account
average, list what you would cut and what you would scale, and
give me your reasoning in two sentences per recommendation.

The creative fatigue check

Look at my top 10 Meta ads by spend over the last 30 days.
Compare each one's CTR and frequency week over week. Tell me
which creatives show fatigue (falling CTR, rising frequency),
which are stable, and which patterns the fatigued ones share.

Fatigue detection is the read-out half of creative work. The metrics worth watching beyond ROAS, hook rate, hold rate, thumbstop, are covered in the creative analytics guide.

The variant batch

Take the brand context for [brand] plus the attached brief from
Notion. Brief the agent to produce 6 new Meta ad variants of our
current best performer: 3 new hooks on the same visual concept,
3 new visuals on the same hook, in 9:16 and 1:1. UGC style, in
[language]. Show me the outputs for review before anything runs.

Know the credit costs before you fire off a batch: a chat message costs 1 credit, a performance report 10, a static image 30, and a 15-second video 500.

Why does creative volume matter for Meta ads in 2026?

Because Meta's delivery system rewards it. Meta Andromeda, the ML retrieval engine behind ad ranking, evaluates far more creative per auction than the old system did, and accounts that feed it diverse creative give it more to select from. The practical consequence: the bottleneck has moved from media buying mechanics to creative production volume, and the teams winning on Meta ship variant batches weekly, not quarterly. The full automation picture is in how to automate Meta ads with AI agents.

That shift is the argument for a full-loop MCP over a reporting-only one. Superscale AI is an autonomous ad agent that researches competitor ads in the Meta Ad Library and TikTok Creative Center, writes scripts and copy, produces video and static ads, resizes them to 9:16, 1:1, and 16:9, publishes to Meta, TikTok, Instagram, and Google Ads, reads performance back, and iterates on the winners. SumUp runs 120+ Meta ads across 8+ languages and 6 product teams on it, and HubSpot CMO Kipp Bodnar called Superscale AI "the best autonomous AI marketing agent that we have seen so far." With the MCP connected, that whole loop is drivable from a Claude chat. Where it sits against the rest of the category is ranked in the best AI marketing agents.

What are the limitations of a Meta ads MCP?

This category is new and the marketing around it runs ahead of the products, so here is the honest list.

  • The data side is plan-gated. Performance reports, competitive research, and budget recommendations need a connected ad account, which requires Advanced ($99/month) or higher. On the free plan and Starter you get creation, not connected-account reporting.
  • Context is per-run. Brand context you pass from Claude, a Notion brief, a reference image, applies to that single run. It is not yet persisted into the stored brand context; persistence is on the roadmap.
  • Parallel runs are capped. The MCP exposes parallel-run slot usage, and large batches queue rather than all running at once.
  • Credits meter everything. 1 per chat message, 10 per report, 30 per static, 500 per 15-second video. A weekly loop is cheap; an unreviewed daily video batch is not.
  • Human review stays in the loop. Publishing to a live ad account from chat is exactly the place to keep an approval step. Ask for outputs, review, then push. The guardrail patterns are covered in our ad creative automation guide.

Frequently asked questions

What is the best MCP for Meta ads? It depends on the job. For reporting and audits only, an open-source Meta ads MCP server or a hosted ad-data MCP is enough. For the full loop, creation, publishing, performance readback, and scheduling in one connector, Superscale AI is the strongest option: its agent generates copy, statics, and UGC video, publishes to Meta, and reads results back into Claude. The free plan includes 1,000 credits with no card required.

Is there an official Meta ads MCP from Meta? No. As of July 2026, Meta has not shipped an official MCP server. Your options are building your own on the Meta Marketing API, running an open-source community server, or using a hosted MCP from an ads platform such as Superscale AI.

Is a Facebook ads MCP different from a Meta ads MCP? No, they are the same thing under two names. Facebook, Instagram, Messenger, and Audience Network ads all run through the same Meta ad system, so one Meta ads MCP covers all of them.

Can Claude create and publish Meta ads through an MCP? Yes, through the Superscale AI MCP. Claude sends a brief to the Superscale agent, which writes the copy, produces static images and UGC video tailored to your stored brand context, and publishes to Meta once you approve. Ask for outputs first and review before anything goes live.

Do I need a paid plan to pull Meta ads performance data into Claude? Yes. Ad-account integrations for Meta, TikTok, and Google unlock on the Advanced plan at $99 per month or higher and are not available on Starter. Ad creation works on any plan, including the free plan with 1,000 credits.

Can I schedule recurring Meta ads reports from Claude? Yes. Any run can be put on a cadence from Claude: a weekly creative refresh, a Monday variation batch, or a Friday performance summary posted to Slack.

Is it safe to connect Claude to my Meta ad account? Your ad account connects to Superscale, not to Claude. Claude talks to the Superscale MCP server, while account access and publishing stay inside Superscale, so there are no Meta credentials sitting in a chat window. Keep a human review step before publishing or scaling, the same rule you would apply to any automation.

How much does a Meta ads MCP cost? Building your own is free apart from developer time. Superscale AI has a free plan with 1,000 credits and no card required, Starter at $49 per month, and Advanced at $99 per month, which is the tier that includes the Meta ad account connection. Credits meter usage: 1 per chat message, 10 per performance report, 30 per static image, and 500 per 15-second video. A 5-day trial adds 3,000 credits.

Run the first loop this week

Connect the account, run the 14-day performance report, ship one six-variant batch, and put the Friday summary on a schedule. That is one working session, not a migration project. Superscale AI starts free with 1,000 credits and no card required, and the 5-day trial adds 3,000 more. If you also buy on Google, set up the sister integration next: Google Ads MCP.

Sources