MCP
Claude is a good marketer with nothing on the desk.
Ask Claude to plan a campaign and it will do a credible job from first principles. Ask it what your category is actually running this month and it has to guess, because nothing in its context tells it.
Connect your own ad accounts and it gets better, but only about you. It can read your spend, your creative, your results, and reason carefully about all of it. What it still can't see is the market those numbers happened in. A drop in performance looks like a creative problem when it might be three competitors entering the auction. A winning angle looks like your idea when half the category shipped it first.
Your own account tells you what happened. It cannot tell you what's possible or what's normal.
Your ads, their ads, everyone's ads.
Marketing context is not one dataset. It's three, and Claude reasons better when it has all of them in the same window.
What you ran, what it cost, what it returned. The baseline you're trying to explain.
Their live creative, claims, offers, and landing pages. The set of moves you know exist.
Advertisers you've never heard of, scaling in your category right now. Where the surprises are.
That third one matters more than teams expect. The most useful creative reference is rarely your named competitor; it's an advertiser you weren't watching who found something that works.
Plug the market in and ask.
MCP is how that context reaches Claude. You connect Upspring once, then ask in plain language, and Claude pulls what it needs mid-thought: advertisers in a category, live ads, creative attributes, scale grades from D to A+.
There's no export, no dashboard to read first, no pre-written summary standing between you and the data. Claude fetches structured context, compares it against your own account, and reaches its own conclusions.
Every ad is a small pipeline. Now multiply it.
This is the part that looks easy from the outside. An ad is not a row in a table. It's a video, a set of images, a block of copy, and a landing page, and its meaning is spread across all four. To make one ad readable by a model, you have to process each of them and then stitch the result back together into something comparable.
Pull frames, transcribe the audio, read the on-screen text, then describe what actually happens in it.
Extract the claim, the offer, the format, the hook, in a vocabulary that stays consistent across advertisers.
Follow the click, render the page, capture the promise and the funnel behind it.
Then you do it again. Campaigns rotate weekly, creative gets swapped without notice, and a library you processed last month is describing ads nobody is running. Freshness is not a step at the end; it's the whole job repeating forever.
And the output has to be small. Handing raw video and full pages to a model burns context and money on every question. Processed creative carries the same meaning at up to 10x fewer tokens, but getting it that small is its own labeling and compression problem, solved once per format and maintained as formats change.
Doing this for a handful of ads is a weekend script. Doing it for a market, continuously, is a system with a team attached, and none of that team is working on your product.
The list you can't escape.
The second problem is quieter and worse. Anything you build yourself starts with a list of advertisers you decided to follow, because you have to point the collection somewhere. That list becomes your ceiling.
You will never discover the competitor you didn't name, the adjacent category eating your audience, or the newcomer outspending everyone this quarter, because your pipeline was never aimed at them. You end up with excellent, expensive data about the market you already assumed you were in.
Coverage first, then processing. That order is the whole difference, and it's why this is infrastructure rather than a script.
What you actually ask.
The questions get bigger once the context is there. Which advertisers in our category are scaling, and what are they claiming. Show me every live ad using a free-trial offer and tell me which ones come from serious spenders. Compare our creative to the top ten advertisers by spend and tell me what we're missing.
Claude already knows how to answer questions like these. Connect the market and it finally has something to answer them with.