InsightsNovember 13, 2026
Meta Ads MCP: Run Meta Ads From Claude
Meta Ads MCP is Meta's own server for AI agents: what Claude or ChatGPT can do in your ad account, how to connect it and the guardrails that stop overspend.
Christoph SchütteMeta Ads MCP is Meta's official Model Context Protocol server, hosted at mcp.facebook.com/ads. It lets an AI agent such as Claude or ChatGPT pull reports, build and edit campaigns, ad sets and ads, and manage catalogs in your Meta ad account. Meta launched it in open beta on April 29, 2026, and new campaigns start paused.
Key takeaways
- Meta's ads MCP server groups its tools into seven categories; in our count of Meta's documentation on November 13, 2026, that is 88 tools, 52 that read and 36 that write.
- Campaigns, ad sets and ads created through the server start paused, and a separate activation tool is the step that starts spending budget.
- You do not need your own Meta app to connect; since July 16, 2026, developers can also connect any AI application through their own developer app.
- Admin rules can stop an agent from creating campaigns or cap budget changes per ad account, but Meta lists them as limited availability.
- The server manages ads, it does not make them: its creative tool builds single-image link creatives from assets already in the account.
What is the Meta Ads MCP?
The Model Context Protocol is the plug between an AI assistant and another system. Anthropic introduced it on November 25, 2024 as "an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools," and in December 2025 the protocol joined the Agentic AI Foundation, a fund under the Linux Foundation. A server publishes tools; an MCP client such as Claude, ChatGPT or Claude Code calls them when you ask in plain language.
Meta's server is the agent half of its ads AI connectors; the other half is an ads command line interface (CLI) for developers. Meta's developer documentation calls it "a Meta-hosted Model Context Protocol server that lets you manage Meta ads directly from an AI agent," one that "works with any MCP-compatible AI agent." The tools are permission-scoped: the agent can only do what the account that signs in is allowed to do.
Meta announced the connectors in open beta on April 29, 2026, with the promise that the MCP needs no developer credentials, API setup or coding. On July 16, 2026, it added a second route for developers:
You can now connect any AI application directly to Meta ads using your own developer app, enabling deeper, more customizable integrations.
Ad buyers were waiting for tools like this for AI agents. In the IAB's 2026 Outlook Study, a survey of 205 US buy-side decision-makers at brands and agencies between November 24, 2025 and January 16, 2026, two-thirds named agentic AI for ad buying and campaign execution as a focus for 2026, and 96% were aware of agentic AI ad buying.
What can the Meta Ads MCP do?
More than reporting. Meta groups the tools into seven categories, but its "Available tools" page was empty when we checked on November 13, 2026, so we counted the tools listed on each category page. Read and write below is our classification from Meta's own tool descriptions.
| Category | Tools | Read | Write | What an agent can do |
|---|---|---|---|---|
| Comprehensive reporting | 7 | 7 | 0 | Pull performance trends, flag anomalies, compare against industry and auction benchmarks, get an opportunity score |
| Ad creation and management | 28 | 17 | 11 | Create campaigns, ad sets and ads (paused), edit and activate them, build single-image creatives, boost Instagram posts, manage custom audiences, search the Ad Library |
| Catalog creation and management | 31 | 15 | 16 | Create catalogs, add and update products, build product sets and feeds, troubleshoot feed issues |
| Signals and datasets | 13 | 7 | 6 | Check pixel and dataset quality, read custom conversions, create pixel event rules that start inactive |
| A/B tests and lift studies | 7 | 4 | 3 | Check eligibility, create, edit or cancel A/B tests, create conversion lift studies |
| Help and troubleshooting | 2 | 2 | 0 | Explain error codes, fetch Help Center articles |
| Activity logs | 1 | 1 | 0 | Read the account's change history, the same view as campaign history in Ads Manager |
| Total (one tool appears in two categories) | 88 | 52 | 36 |
Two gaps stand out. There is no tool to delete a campaign, ad set, ad or creative; deletes exist only for custom audiences, catalog products, product sets, feeds and pixel rules. And Meta publishes no numeric rate limit: its rules guide only tells developers to watch the usage header and back off as they approach the limit.
The reporting half is where most teams start, and the tools go further than raw spend: one flags anomalies, two compare your results with industry and auction benchmarks, one scores opportunities. We hear the same order in our sales calls. A performance marketer we spoke with this year had already connected Claude to Meta for a monthly report; the useful next step was cadence, not more data: pull last week's winners every Monday and ask for new briefs from the concepts that worked. Before acting on any return the agent reports, check it against your break-even with our ROAS calculator.
How do you connect Claude or ChatGPT to Meta Ads?
Meta's get-started guide says owning a Meta app is not a prerequisite. With an app, you add the use case "Create & manage ads with ads MCP server" to it in the Meta developer dashboard. Either way, the setup comes down to five decisions:
- Choose who the agent acts as. With OAuth, your AI client opens the Facebook Login for Business dialog and you sign in with your Facebook account or a Meta Managed Account. For automation, Meta recommends a system user, since it "is not tied to a person who can leave the business or lose access." Only Employee-role system user tokens work; Admin-role tokens are refused.
- Grant the smallest scope. The server needs ads_mcp_management plus either ads_read, for reporting only, or ads_management, to build and edit. Catalog, business, Pages and Instagram permissions add further tools.
- Add the server to your client. In Claude Code, add an HTTP server with the address mcp.facebook.com/ads and your Meta app ID as the client ID. In ChatGPT, turn on developer mode, choose OAuth as the authentication method, enter the same server address and your app ID as the OAuth client ID.
- Pick the ad accounts. The sign-in only reaches the accounts that person or system user can access in the business portfolio, so agencies connect client accounts the way they already share them.
- Test read-only first. Ask for spend, results and cost per result by campaign for the last seven days and compare the numbers with Ads Manager before you grant write access.
Can an AI agent spend your budget?
An agent with write access can spend real money, but only through one deliberate step. Meta's documentation of the ad creation tools says: "Write tools create entities in a paused state; your AI client asks for confirmation before activation." The tool that flips a paused campaign, ad set or ad to active is described in its own words as the one that "starts spending budget."
The second layer is rules. Meta's rules guide lets an admin deny an agent specific actions on one ad account or catalog: creating campaigns, ad sets or ads; budget edits above a percentage, an amount or an absolute maximum, for daily, lifetime or both budgets; edits to targeting, creative or status; or every action at once. "Rules are the same controls shown on the ads MCP server page in Meta Business Suite settings." They are in limited availability, and a business that is not enrolled gets an error when it tries to set one.
A sensible order for a live account:
- Weeks one and two: ads_read only. Let the agent report, and compare its numbers with Ads Manager.
- Before write access: set a budget rule, such as a 20% cap on any single budget change, if your business is enrolled.
- With write access: keep the approval step human. Every new campaign arrives paused, and nothing goes live until you confirm; review it in Ads Manager before you do.
- Every week: read the activity log through the agent and ask it to list every change it made.
Official server or a third-party Meta Ads MCP?
Before Meta shipped its own, advertisers used community servers built on Meta's Marketing API, and they still exist. Pipeboard's meta-ads-mcp, a popular project on GitHub with about 1,400 stars when we checked, calls itself "an independent open-source project that uses Meta's public APIs." It is licensed under the Business Source License 1.1 until it becomes Apache 2.0 on January 1, 2029, and runs as a hosted service by a badged Meta Business Partner.
| Meta's ads MCP server | Third-party server (example: Pipeboard) | |
|---|---|---|
| Who runs it | Meta, at mcp.facebook.com/ads | The vendor, or you when self-hosted |
| Sign-in | Facebook Login for Business, or your own developer app | Vendor sign-in, or your own Meta app when self-hosted |
| Tools | 88 in seven categories (our count) | 28 in the README's tool list |
| New campaigns | Start paused, confirmation before activation | Start paused, confirmation on every write (README) |
| Admin rules | Yes, limited availability | Meta's MCP rules do not apply |
| Price | None stated on Meta's pages | Free plan, then paid tiers |
| Creative upload | No upload tool | Listed in the README |
For anything that writes to a live account, Meta's own server is the safer default: it runs on Meta's permission model and its admin rules apply. A third-party server earns its place where it covers something Meta's does not yet, such as cross-platform reporting. For scripts, Meta's Ads CLI covers create, read, update and delete for campaigns, creatives and catalogs with a system user token, including image upload from a file.
Where does creative fit in an MCP workflow?
The MCP can tell you what to make; it cannot make it. Its creative tool builds "a single-image link ad creative," combinations of images and video are in limited availability, and there is no upload tool: the image and video tools only list what is already in the account. An agent can search the Ad Library, read your previews and see which concepts won, then hand you a brief.
That puts the pressure where it already was. Meta's delivery now rewards distinct concepts over near-copies, as our explainer on Meta Andromeda shows, so an agent that reads results every week will ask for new ideas every week. In the calls where teams describe this loop, the same gap comes up: the analysis now takes minutes, and production is what sets the pace. Our guide to AI video ads covers what works on that side.
How to close the loop with Solid
Solid, the AI creative engine for brands and their agencies, has its own MCP connector, so the chat that reads your Meta results can also work on the creative. Connected to Claude, ChatGPT, Grok or Langdock, it searches footage and scenes in your Smart Library, creates video projects, folders and tags, and reads or updates your brand guidelines. In Solid, the Solid Agent turns the brief into finished ads from your real footage, and you export every format for upload in Ads Manager. Setup is one connector address and a sign-in; see Solid's MCP connector for Claude and ChatGPT.
FAQ
Is the Meta Ads MCP free?
Meta's announcement and developer documentation name no price; the connectors are in open beta. You pay for ad spend as usual, and your AI client may need a paid plan that allows custom connectors. Third-party servers set their own prices; Pipeboard, for example, has a free plan and paid tiers. If Meta introduces pricing after the beta, it would appear on the same documentation pages.
Which AI tools work with the Meta Ads MCP?
Meta says the server works with any MCP-compatible AI agent, and its setup guide gives steps for Claude Code and for ChatGPT in developer mode. Other clients that support remote MCP servers with OAuth sign-in should connect the same way. A quick test: if your client can add a custom connector by URL and send you through a sign-in page, it can reach mcp.facebook.com/ads.
Can the Meta Ads MCP delete campaigns or ads?
No. When we checked, the server had no delete tool for campaigns, ad sets, ads or creatives, only an update tool and an activation tool. An agent can pause or edit, and it can delete custom audiences, catalog products, product sets, feeds and pixel rules. Deleting a campaign still happens in Ads Manager or through the Ads CLI.
Ready to make what your agent asks for?
Pair Meta's MCP with Solid and turn last week's winners into new concepts from the footage you already own.
Start free trial or Book a demo
About the author
Christoph Schütte is Co-Founder and CTO of Solid, the AI creative engine for brands and their agencies. A physicist and former Engineering Lead at Google, he worked on video intelligence and computer vision in Google's self-driving cars and robots unit and with brands at Google Shopping on the loop between ad performance and production. Connect on LinkedIn
Sources
- Meta for Business: Introducing Meta ads AI connectors, April 29, 2026, with the update of July 16, 2026
- Meta for Developers: ads MCP server overview, get started, ad creation and management tools, the six other tool category pages, rules best practices and Ads CLI overview, checked November 13, 2026
- Anthropic: Introducing the Model Context Protocol, November 25, 2024
- Model Context Protocol: MCP joins the Agentic AI Foundation, December 9, 2025
- IAB: 2026 Outlook Study, January 2026
- Pipeboard: meta-ads-mcp on GitHub, README and license, checked November 13, 2026