TL;DR: There is no official Meta MCP server. Every option is third-party, which means your campaign data flows through infrastructure you don't control and setup requires a Meta Marketing API token. Some servers support write operations, meaning an AI can pause campaigns or change budgets on your live account. Be careful with that.
Reporting in Meta Ads Manager is slow. Find the right breakdown, filter by date, maybe export to a spreadsheet. Do that across multiple campaigns, ad sets, and creatives, and you spend more time pulling reports than acting on them.
A Meta Ads MCP server connects AI tools like Claude directly to your ad account. You ask a question in plain English, it queries your live campaign data through the Meta Marketing API and gives you a structured answer.
The catch: Meta hasn't released an official MCP server. Every option available today is built by a third party. That is a meaningful difference from Google Ads MCP, where Google's own Ads API team maintains the server. With Meta, you are trusting someone else's code with access to your ad account.
What is Meta Ads MCP?
Meta Ads MCP is a server that connects AI tools like Claude or Cursor to your Facebook and Instagram ad account using the Model Context Protocol standard. It sits between your AI tool and the Meta Marketing API, translating plain-English questions into API calls that fetch (and sometimes modify) your campaign data. Instead of writing code or clicking through Ads Manager, you ask a question and get a structured answer from live data.
What you can do with a Facebook Ads MCP server
Once connected, you interact with your campaigns through conversation instead of clicking through Ads Manager.
Campaign performance
Ask which campaigns drove the most conversions last week, which ad sets burned through budget with nothing to show for it, or which placements are dragging your CPM up. Instead of building a custom report in Ads Manager, you describe what you want and get a structured answer from live data.
Creative comparison
Ask which images or videos are driving the best conversion rate across your ad sets, or which format (single image vs. video vs. carousel) has the lowest CPC. This kind of cross-creative analysis is tedious in the Ads Manager UI since you have to nagivate into each ad set individually. With MCP, it's one question.
Creative fatigue detection
Creative fatigue happens when your audience has seen your ads too many times and performance drops. MCP can flag ads with declining CTR or rising frequency, so you know which creatives to refresh before ROAS tanks.
Audience insights
Which audiences have the lowest cost per acquisition? Which retargeting lists are converting and which are exhausted? Audience analysis in Meta involves a lot of clicking between ad sets. MCP gives you a cross-campaign view in one go.
Budget pacing
Which campaigns are underspending their daily budget? Which ones hit their limit by noon and miss afternoon traffic? Easy to miss in a large account, easy to surface with MCP.
Cross-campaign reporting
Compare performance across campaigns in a single answer. Ask for your top 5 campaigns by ROI, or find every ad set where CPM is above $20 and CTR is below 1%. The kind of multi-filter analysis that takes several minutes to set up manually.
Write operations (careful)
Some MCP servers (Pipeboard and brijr/meta-mcp) support write operations. That means an AI agent can pause campaigns, adjust budgets, update bids, or create new ads on your behalf.
This is useful, and also dangerous. An AI misunderstanding "pause the underperforming campaigns" could mean pausing the wrong ones. On an account spending $10k/month, that is expensive. If you use write access, configure confirmation steps, start in a test environment, and never hand over unreviewed changes to a high-spend campaign. This kind of marketing automation requires guardrails that most MCP servers do not yet provide.
Meta Ads MCP servers compared
There is no single "Meta Ads MCP." You choose from several third-party servers:
| Pipeboard | GoMarble | brijr/meta-mcp | Composio | |
|---|---|---|---|---|
| Maintained by | Pipeboard.co | GoMarble AI | Community (brijr) | Composio (deprecating) |
| Cost | Free tier + paid | Free (open-source) | Free (open-source) | Free (deprecating) |
| Setup difficulty | Low (hosted) or Moderate (self-hosted) | Low (automated installer) | Moderate (self-hosted) | Low |
| Read/Write | Read + Write | Read-only | Read + Write | Read + Write |
| Tools count | 26 | 25+ | 25 | 53+ |
| AI tool support | Claude, Cursor | Claude, Cursor | Any MCP client | Any MCP client |
| Best for | Most marketers (hosted, most complete) | Read-only analytics | Developers wanting full control | Not recommended |
| License | BSL 1.1 (Apache 2.0 Jan 2029) | MIT | MIT | Open source |
Source: GitHub repos and product pages, verified March 2026.
Which one to pick?
- Pipeboard: easiest setup, most complete feature set. The hosted option skips server configuration entirely. BSL 1.1 license means you can't build a competing commercial service on top of it, fine for internal use.
- GoMarble: read-only, fully open-source (MIT). Good if you just need analytics and want to audit every line of code.
- brijr/meta-mcp: self-hosted, full read/write, works with any MCP client. For developers who want control.
- Composio: deprecating their MCP product. Don't start with this.
How to set up Meta Ads MCP
Three steps. Step 1 is where most people get stuck.
Step 1: Get a Meta Marketing API access token
You need a Marketing API access token from Meta. This requires a Meta developer account and a registered app with Marketing API access. The Meta Marketing API docs walk through the process, but it's not as quick as OAuth flows you might be used to. Expect 15-30 minutes if you haven't done it before. You'll also need your Ad Account ID from Meta's Business Settings.
One thing to watch: token expiration. Short-lived tokens expire in hours. System user tokens last longer but require a Business Manager setup. If your MCP connection suddenly stops working, an expired token is almost always why.
Note: MCP connects to your ad data, not your tracking. If you also need to set up conversion tracking, see our Meta Pixel installation guides.
Step 2: Choose your MCP server
Pick one from the comparison table above. For most people, Pipeboard's hosted option is the lowest-friction path since you don't need to run anything locally.
If you go self-hosted, you'll need Node.js installed. The setup is similar across servers: clone or install the package, configure it with your API token and Ad Account ID.
Step 3: Configure your AI tool
For Claude Desktop, add the MCP server to your configuration file (claude_desktop_config.json). Claude will then have access to your Meta ad data in any conversation.
For Cursor or other MCP-compatible tools, the process is similar: add the server to the tool's MCP settings.
After setup, test with something simple: "What campaigns are currently active in my account?" If you get real data back, you're connected.
Limitations and risks of Meta Ads MCP
No official server
This is the big one. Google has an official MCP server maintained by the team that builds the API. Meta has nothing. You're relying on third parties to keep up with Marketing API changes. If Meta ships a breaking update, there's no guarantee your server updates in time, or at all.
These servers have full access to your ad account
Read that again. When you give a third-party MCP server your Marketing API token, you're handing over the keys. Pipeboard, GoMarble, brijr: they can see your campaign structure, your budgets, your targeting, your creative performance, your audience data. The ones with write access can modify all of it.
GoMarble and brijr are open-source (MIT), so you can audit the code. Pipeboard is BSL-licensed, you can read it, but you can't fork and commercialize it. But even with open-source code, you're trusting that the version running on a hosted service matches what's on GitHub. You don't know what they log, what they store, or how long they keep it.
If you're connecting a tool to an account that spends real money, this should make you uncomfortable. At minimum, read their privacy policy. Better yet, self-host so nothing leaves your machine.
Write operations on live campaigns
Some of these servers can pause your campaigns, change your budgets, create new ads. Through code you didn't write, maintained by people you don't know. On your live account, with your real ad spend.
There are no undo buttons in Meta Ads Manager. A bad prompt, a misinterpretation, a bug in the server code, that's money gone. On a $10k/month account, "pause the underperforming campaigns" misunderstood could mean your best campaign goes dark for a weekend.
If you use write access, use confirmation steps, set spending caps, and never give unreviewed control to high-spend campaigns.
Rate limits
The Meta Marketing API enforces rate limits (roughly 200 calls per hour per ad account). Detailed analyses on large accounts can hit this cieling regardless of which server you use.
Token management is manual
Unlike some OAuth flows that refresh automatically, Meta API tokens need to be renewed. Set a reminder to check token expiration, or you'll lose your MCP connection without warning.
AI can misread the data
The MCP server queries real numbers, but the AI's interpretation isn't perfect. If you're making a significant budget decision based on what the AI reports, verify the numbers in Ads Manager first. Use MCP to narrow your focus, not as your only source of truth.
Meta Ads MCP vs Google Ads MCP
If you run campaigns on both platforms:
| Meta Ads MCP | Google Ads MCP | |
|---|---|---|
| Official server | No (third-party only) | Yes (Google Ads API team) |
| Read/Write | Read + Write (some servers) | Read-only (official) |
| Setup difficulty | Moderate (API token required) | Moderate (OAuth + API credentials) |
| Cost | Free tier available | Free (open-source) |
| Reliability | Varies by server | Maintained by Google |
Since both use the MCP standard, you can connect Meta and Google Ads to the same AI tool. Ask something like "Which platform had a better ROAS last month?" and get a real answer from live data across both accounts. That cross-platform reporting works today, even if setup takes some work.
The Google Ads MCP guide covers the Google side in detail.
What's coming
MCP adoption across ad platforms is accelerating. Google has an official server. Amazon Ads launched theirs in early 2026. An official Meta MCP server feels likely given how much ad spend runs through Facebook and Instagram, but Meta hasn't announced anything.
If that happens, the third-party options covered here may become less relevant. Until then, Pipeboard and GoMarble are the practical choices.
For AI-driven media buying to work properly, write access needs better safeguards: confirmation flows, rollback mechanisms, spending caps. Right now, the risk of AI making mistakes on live campaigns is high enough that most people should stick to read-only analysis.
At AdKit, we are building toward MCP-powered ad analysis across platforms. The goal is cross-platform campaign intelligence without setting up separate MCP servers for every ad platform. One interface for Google, Meta, and beyond. The broader shift is toward AI ad agents that don't just retrieve data but understand context and surface insights you wouldn't have looked for.
While you set up your MCP connection, you can use our free ROAS calculator and ad profit calculator to benchmark your current campaign performance.
