ChatGPT for Meta Ads: what works through MCP

OpenAI documents full custom MCP support, including write actions, for Business, Enterprise and Edu, with Pro limited to read/fetch permissions. These custom apps are available on the web. Workspace controls determine who can create and use them. Check the linked OpenAI page if your settings differ; availability is still described as beta.

A read-only plan does not turn an Adrails proposal tool into a read. Preparing a budget change records a pending action, so it is a mutation even though no advertising change has been applied. Start with reporting when the client or workspace does not allow that tool.

Connect your Meta ad account in Adrails first. In ChatGPT web, use your workspace's custom MCP app controls to add https://adrails.ai/api/mcp with OAuth. Sign in to Adrails, select the intended workspace and grant the permissions needed for the task. Review the discovered tools before using the app in a conversation.

Begin with an account list and a read request. Ask the assistant to identify the account and report the date range and currency with the figures. A successful sign-in does not mean every Meta account you can access is connected to this Adrails workspace.

performance_period reads Meta results over an exact period and can compare another period. campaigns_structure shows the campaign, ad set and ad structure from the last synchronization. creative_insights groups ads by their creative to help inspect performance and fatigue. These tools supply account evidence; the explanation still needs to distinguish an observed change from its possible cause.

A useful prompt names the decision: compare campaign CPA with the previous period, inspect whether delivery changed, and list what evidence is missing. Ask the assistant to check snapshot freshness before quoting saved structure or analysis. Do not treat a confident explanation as proof of why delivery changed.

• Compare this month's Meta campaign results with last month and show the account, currency and reporting periods.
• Show the current campaign structure and say when the saved account data was synchronized.
• Review creative performance and separate signs of fatigue from missing evidence.

With the necessary client access and Adrails permission, campaigns_change_budget prepares a budget change, campaigns_set_status prepares a pause or resume, and campaigns_launch prepares a Meta campaign build from a completed brief. The advertising adapter returns a confirmation link and a pending result. It does not apply the advertising change.

Open the link and review the target account, affected entities and exact proposed change. An owner or admin applies it in Adrails, subject to the plan and action checks. New launches are created paused; enabling spend is a separate action. Allowing a tool call in ChatGPT is not a substitute for this confirmation in Adrails.

Use Meta's own tools for billing, account administration and settings that Adrails does not expose. MCP is a way to call supported operations, not unrestricted control of Ads Manager. If a tool is absent, the assistant should say what it cannot do instead of inventing a successful result.

Advertising proposals also differ from automation management. A separately authorized workflow can deliver messages or execute its configured actions; it is not covered by the blanket claim that every MCP mutation only prepares an advertising proposal. For a first campaign review, keep the granted permissions tied to reporting and advertising preparation.