AI Agent for ChatGPT Ads That Can Make Changes

How PaidSync.ai keeps you in control

• New ad groups start paused. Nothing new serves until you enable it.
• A confirm for risky moves. Pausing, enabling and archiving refuse to run without an explicit confirm, and archiving is permanent, so it says so before it runs.
• Approval before each change. In the PaidSync.ai workspace, every change waits as an approval card until a person approves it. From Claude, your assistant's own approval setting decides whether it asks first.
• Your key stays yours. The OpenAI Ads API key is encrypted at rest with AES-256-GCM, and disconnecting deletes it.
• A change log. Live changes are recorded with the user, the tool and the account active in the session.

An MCP server turns an ad platform's API into typed tools an AI assistant can call. Connect one to Claude and "pause every ChatGPT Ads ad group whose CPC is above $2 in the last 14 days" becomes something the assistant can check, propose, and, once you approve, execute. In ChatGPT, PaidSync.ai reports on ChatGPT Ads today. This page covers PaidSync's ChatGPT Ads surface in full: the tools, exactly what they do and do not do, how writes are checked, and how it compares with the ChatGPT Ads MCP options on most current lists.

It is written for both audiences. Marketers who want to manage ChatGPT Ads campaigns day-to-day from a chat window, and agent builders who want to know exactly what the protocol surface looks like before wiring it into a workflow. Each section is labeled.

What an AI agent for ChatGPT Ads does

Three things, in practice. It builds, so a campaign, its ad groups, its creative and its ads can be assembled from a brief in conversation. It reads, so the assistant can pull spend, clicks and performance without opening OpenAI Ads Manager. And it gates, so creates and updates preview as a dry run first and destructive changes need an explicit confirm.

OpenAI began serving ads in ChatGPT in February 2026 and opened self-serve buying in May. PaidSync is an OpenAI Select Partner and the ChatGPT Ads surface here is a full build tier, the same tier as Google Ads, Meta, LinkedIn, TikTok and Microsoft Ads: create from scratch, then manage daily. That is different from the manage-only platforms such as TikTok, Microsoft, Snapchat, Reddit, Pinterest and X, where campaigns are built in the native manager first. The capability matrix for all 14 platforms is on the full MCP server page.

Every ChatGPT Ads tool, by category

PaidSync.ai's live server exposes the ChatGPT Ads tools below, by category (server total 630+ across 14 platforms). The table names all of them; the live catalog is also discoverable from any session via paidsync_context with include_tools:['openai'].

The build path is the part worth understanding. create_openai_campaign opens the campaign with its objective and budget, create_openai_adgroup sets targeting and bids beneath it, upload_openai_creative_image puts the image in place, and create_openai_ad assembles the chat_card ad itself. Each of those four is a separate call, and the three create steps preview as a dry run first, so a full build is deliberate rather than one blind click. After launch, get_openai_performance and the list_* readers feed the weekly pass, and the pause, enable, archive and update tools carry it out.

Manage ChatGPT Ads with an AI agent

The practical shape of a week on PaidSync's ChatGPT Ads stack looks like this. Monday morning you ask "what changed in the last 7 days on ChatGPT Ads, by campaign, and where is CPC drifting". The assistant pulls it from get_openai_performance and answers in plain numbers. You spot two ad groups pacing over budget on stale creative, ask for a pause on the worst offender and a 30% budget cut on the other, review the drafted changes, and confirm. Later you ask for a status report on live campaigns; the assistant lists them from list_openai_campaigns.

Because the same server carries Google Ads, Meta Ads, LinkedIn, TikTok, GA4, and GTM, the follow-up question "did the ChatGPT Ads cut shift volume to Meta shopping and did GA4 see the difference" is answered in the same session, not in another tab. That cross-platform session is the practical difference between a ChatGPT Ads connector and a paid-media server.

Trust is procedural, not promissory. Over MCP, the confirmation step is your assistant's own prompt; PaidSync writes no approval card on that path. Previews show the exact planned change first on creates and updates, and destructive operations require an explicit confirm_destructive flag. A connected sign-in lasts 30 days at most, then ChatGPT asks you to sign in again. You can also end every assistant sign-in from Settings; apps on an API key keep working until you rotate the key.

The protocol surface behind the tools

PaidSync is one MCP endpoint covering 14 platforms; details on the MCP server page. To keep client context small, tools/list returns a lean starter set. The full catalog is discovered at runtime through paidsync_context, which returns the active account, connected platforms, and the grouped tool catalog (include_tools:['openai'] returns the exact ChatGPT Ads tool names), and executed through paidsync_exec. That two-call pattern means an agent can enumerate and invoke every ChatGPT Ads tool without a 630+ tool schema payload sitting in every conversation.

Safety is enforced server-side. Create and update tools accept dry_run for a no-op preview, and destructive operations require confirm_destructive. For discovery and verification, the domain publishes OAuth authorization-server and protected-resource metadata (RFC 8414 and RFC 9728), an MCP server card, agent-auth.json, security.txt, and a Content-Signal directive.

The ChatGPT Ads MCP landscape

What exists for OpenAI's ads product falls into two groups. Open-source community servers wrap the OpenAI Ads Advertiser API and, in their first releases, read only: campaigns, ad groups, ads and insights, with writes on the roadmap. Single-network hosted MCPs go further and add writes, but they cover ChatGPT Ads alone, so a question that crosses into Google or Meta leaves the session.

PaidSync sits in a third place. Full build and manage on ChatGPT Ads, dry runs before creates and updates, and 13 other platforms in the same conversation, which is what makes "shift the budget that ChatGPT Ads is not spending into Meta prospecting, and show me GA4 after" a single thread instead of three tools.

What this does not do

Stated plainly, because scoping a tool accurately beats overselling it.

• No video creative. Image creative uploads through upload_openai_creative_image; video is not exposed as a tool today.
• No audience management tools. Audience setup stays in OpenAI Ads Manager.
• No billing or payment changes. Budgets yes, payment methods never.
• One object per task. Building a campaign with three ad groups is four separate tasks, not one.
• No blind writes. Creates and updates preview as a dry run first, and over MCP your assistant's own prompt is the human check.

Roadmap decisions on ChatGPT Ads coverage track customer demand. If a capability above is central to your workflow, tell us at the demo booking page.

Pricing

Every plan includes all 14 platforms and there is no per-account or per-platform charge. Every tool call counts as one task. Full detail at paidsync.ai/pricing.

Frequently asked questions