The most interesting media buying operations in 2026 are one person and a chat window. Not because AI replaced the buyer — because MCP replaced the tabs. The AI media buyer stack in 2026 is an assistant (Claude, ChatGPT, Claude Code, or Cursor) with a handful of MCP servers plugged into it, each covering a layer of the job that used to be its own subscription, its own login, and its own afternoon.
This post is an honest survey of that stack: what the layers are, who actually covers each one, where AdWhispr sits, where Adspirer sits, where the spy tools sit — and the pricing consideration that matters more than any feature list.
The three layers of the job
Strip media buying to its verbs and there are three:
- Research — what's working in my category? What are competitors running, and what's proven vs merely live?
- Creative — turn that intelligence into ads for my brand.
- Execution — launch, budget, pause, scale. The account-touching work.
The 2026 question for any tool is simple: which layers does it cover, and can your assistant drive it directly over MCP?
| Layers covered | MCP-native? | What it can't do | |
|---|---|---|---|
| Spy tools (Foreplay, Atria, MagicBrief, AdSpy, BigSpy) | Research only | Mostly no | Can't clone for your brand, can't launch anything |
| Adspirer | Execution + management | Yes | No competitor ad library, no creative engine |
| AdWhispr | Research + creative + execution | Yes | No LinkedIn or Amazon execution; Meta execution still beta |
Let's take them in order.
The research layer: spy tools
The legacy answer to layer one is a spy tool — a searchable feed of competitor ads you browse and save. They're fine as swipe files. The structural problems are two: the output is a screenshot (you still rebuild everything by hand), and most of them print performance metrics that don't exist. Meta's Ad Library exposes no CTR, CPC, or ROAS — so any "ROAS score" on a competitor ad is invented. We've made the case against fabricated metrics at length; the honest alternative is days-running, the one signal you can verify, because brands don't keep paying for losing ads.
In a chat-first stack, a research tool your assistant can't call is a research tool you'll stop opening.
The execution layer: Adspirer
Adspirer is the best-known execution MCP server, and credit where due: it's a serious product. 340+ tools spanning Google, Meta, Amazon, LinkedIn, TikTok, and ChatGPT Ads. A sane safety model — campaigns are created paused, and there are no destructive tools. Watch Agent monitoring. Fast shipping. The category itself is clearly growing — Adspirer's publicly verified revenue sits around $55k MRR on TrustMRR as of June 2026.
If your job is operating many accounts across many platforms — especially LinkedIn or Amazon, which AdWhispr doesn't execute on — Adspirer belongs in your stack.
Its boundary is equally clear, and it's by design: Adspirer manages the campaigns you already have. It has no competitor ad library, no competitor intelligence, and no creative generation — you bring the strategy and creative direction separately. Layers one and two are someone else's problem. We compared the two products in depth in AdWhispr vs Adspirer.
The full-loop layer: AdWhispr
AdWhispr started as layer one — competitor ad research over MCP, with a moat built on verifiable signals: full Meta Ad Library ingestion per brand, TikTok ad research, Google keyword intel, AI classification of every ad (hook, format, tone, offer), and days-running as the performance proxy. At the time of writing that's 380+ tracked brands and roughly 194,000 ads, 74,000+ of them running 100+ days.
Then it grew down the stack. Layer two: clone_ad and clone_tiktok_ad rebuild a competitor's proven ad for your brand, image or video. And as of June 2026, layer three: connect_ad_account (OAuth), then launch_search_campaign and launch_pmax_campaign on Google (GA), launch_tiktok_campaign on TikTok (GA), launch_cloned_ad to ship a clone directly, plus management tools (list_campaigns, get_account_performance, pause_campaign, update_budget). Meta execution is in beta.
That's the "layer above" position: Adspirer executes what you bring it; AdWhispr tells you what to run — pulls the competitor's proven winners, clones them, and launches them, same chat, one prompt. The whole loop is one sentence:
"Find the longest-running ad from @competitor, clone it for my brand,
and launch it on TikTok with a $50/day budget"
We wrote up the one-prompt loop as its own post, plus tutorials for connecting your ad account to Claude or ChatGPT and the clone-and-launch workflow step by step.
The pricing consideration: flat vs metered
Here's the stack decision nobody puts on the feature grid, and it changes how you behave in the chat.
Adspirer meters per tool call: free tier is a one-time 15-call snapshot, Plus is $49/mo for 150 calls with $0.50-per-call overage, Pro is $99/mo for 600, Max is $199/mo for 3,000, Enterprise from $18k/yr. For scheduled account operations that's workable — you can budget calls like API quota.
But research doesn't work on a meter. Good competitor research is twenty questions where the nineteenth is the one that matters. When every "wait, also show me..." has a marginal price, you start rationing curiosity — which defeats the point of having an analyst in the chat.
AdWhispr's pricing is flat by design: Free (5 tool calls/mo, 1 brand) to try it; Pro at $39/mo with unlimited research tool calls, 3 brands, and 10 clones a month (3-day free trial); Agency at $149/mo with 10+ brands, 50 clones, and cross-brand comparison. Clone credit packs (10/$19, 30/$39, 100/$99) cover heavy creative months without a plan jump. Ask your hundredth research question at the same price as your first.
The one-person stack, assembled
So what does the solo operation actually run in 2026?
- An assistant — Claude, ChatGPT, Claude Code, or Cursor. The cockpit.
- AdWhispr for research, creative, and launch on Google Search/PMax and TikTok — the strategy-to-live-campaign loop. Setup at adwhispr.com/integrations.
- Adspirer if you need execution breadth AdWhispr doesn't cover — LinkedIn, Amazon, or heavy multi-account operations across platforms.
- Spy tools — honestly, only if you're attached to the swipe-file habit. The research layer is better served by tools your assistant can query directly, with signals it can verify.
One person, one chat, and the three layers of the job answering in the same conversation. That's the stack.
Start free at adwhispr.com — 5 tool calls is enough to see your first competitor's proven winner.