This marketing API is built to be asked, not just read. Connect it to Claude, ChatGPT's developer mode, Cursor or Gemini CLI, and you can ask it plain questions about pacing, anomalies, performance by campaign or client, and get an answer pulled from your real, current data — not a guess the model makes up because it can't see your account. You choose which clients an assistant is allowed to touch when you connect it, so an agency managing forty accounts doesn't have to hand an AI tool the keys to all forty just to ask about one.
Ads and Pixels / MCP & API
A marketing API that never approves its own changes
A remote MCP server for Claude, ChatGPT, Cursor and Gemini CLI, plus a REST API and signed webhooks on the same service layer. An AI assistant can read everything you allow and propose changes — approving is a human action, every time.
One place. Every screen.Explore below
01 /
Ask your assistant, and it actually knows the answer
02 /
It can propose a change. It can never approve one.
Ask an assistant to pause a losing ad set or shift a budget, and it can propose exactly that — the proposal lands in the same approval queue a human already uses for every other change, checked against the same guardrails the AI optimizer runs under. Approving is never something an AI tool can do here, by design, in every scope this marketing API offers. A person in the app clicks approve, every time, no matter how the change was proposed.
03 /
Write access only where autopilot already allows it
A write scope exists, but it only does anything on a client where autopilot is already turned on for that action type, and it still passes the same limits as the AI optimizer — at most a 20% budget or bid change per step, at least 72 hours between changes, never touching a conversion event or attribution setting. This marketing API doesn't create a second, looser door into your accounts; it's the same guardrails, reached a different way.
04 /
Ad copy and search terms are treated as untrusted data
Ad names, creative text and search terms that come back through this API are handled as untrusted data, not instructions — because a malicious or careless piece of text inside your own ad account shouldn't be able to trick an AI assistant into doing something you didn't ask for. That's a real risk with any tool that lets a language model read live account data, and it's treated as one here rather than assumed away.
05 /
The same access, as a REST API and webhooks, for your own code
A marketing API worth using shouldn't lock you into a chat interface. Everything the MCP server can do is also available as a documented REST API (OpenAPI spec) with per-agency API keys, so a developer building an internal tool isn't limited to talking through an AI assistant. Signed webhooks push events as they happen — a change requested, approved, applied or failed; a spend cap reached; an anomaly detected; a report ready — so your own systems can react without polling.
06 /
Scoped, rate-limited, and logged
Every connection is scoped to exactly what it needs — read clients, read reports, propose changes, or write within autopilot — and nothing more, following the same least-access principle any good marketing API should. Rate limits apply per connection and per agency, so one assistant or integration can't burn through another client's quota to a network. Every call and every proposed or applied change is written to an append-only audit log, visible from inside the app, so there's always a record of what asked for what and who approved it.
07 /
What this doesn't do
- It doesn't let an AI assistant approve its own proposal, under any scope, on any plan — that step is reserved for a human in the app.
- It doesn't hand over every client the moment you connect an assistant; you choose which ones it can see.
- And it doesn't treat creative text or search terms from your own account as trustworthy instructions just because they're already inside your data.
A little more clarity
Frequently asked questions
Real answers before you make a decision.
Which AI assistants work with this marketing API?
Claude, ChatGPT's developer mode, Cursor and Gemini CLI connect through the remote MCP server today; anything that speaks the standard MCP protocol over Streamable HTTP can connect the same way.
Can an AI assistant spend my budget through this?
Only if you've turned on an autopilot tier for that specific client and action, and only within the same fixed limits the AI optimizer enforces everywhere else. Approval for anything outside that is always a human action in the app.
Is there a plain REST API, or do I have to use an AI assistant?
Both are available on the same service layer, with per-agency API keys for direct REST access and an OpenAPI spec for your own integrations, plus signed webhooks for real-time events.
How does this protect against prompt injection from ad data?
Ad names, creative text and search terms pulled through the API are returned as labeled, untrusted data, never as instructions an assistant should follow — a defense built in from the start, not bolted on after an incident.
Can I limit which clients an assistant is allowed to see?
Yes. You choose the specific clients an assistant can touch when you connect it, so it never has automatic access to your whole account list — the same scoping applies whether you're using the MCP server or the raw marketing API directly.
What webhook events can my own systems subscribe to?
Change requested, approved, applied or failed; spend cap reached; anomaly detected; and report ready — all signed, so you can verify they actually came from us.
Does this cover email or social media marketing API access too, or just paid ads?
Paid ad campaign, performance and pacing data is the core of it today. It isn't an email marketing api or a general social media marketing api for organic posting — those are outside its scope.
Is this a digital marketing api for developers, or only accessible through an AI assistant?
It's a digital marketing api built for developers first — full REST access and webhooks — with the MCP/AI-assistant layer as one more way in, not the only one. Marketing api developers get the same OpenAPI spec whether or not they ever touch an AI assistant.
Is there an affiliate marketing api for tracking affiliate or partner sales?
Not directly — this covers ad campaign management and performance, not an affiliate marketing api for partner/referral tracking specifically.
Your next move
Connect your assistant, keep the approval human
Set up the MCP server or grab an API key, choose which clients it can see, and start asking real questions about real data.