Pillar · MCP · for multi-account PPC
The MCP server that turns Claude Code into a Google Ads decision engine.
You already paste exports into Claude, or you've tried Google's Ads MCP. Both stop at your data. AdTAO MCP adds what neither can: 20,000+ accounts of peer context, pattern-level account intelligence, and predict-before-you-act grounded in real outcomes — with every write staged for your approval. Wrapper MCPs give you more GAQL. We give you decisions you couldn't make alone.
Connected advertiser corpus
20,000+
peer context your CSV can't include
Scenarios published
5
adversarial traps · same prompts for all setups
Safe on "change now" prompts
100%
staged · approve-then-execute · in the published run
Median time to answer
~3.5 s
composed server-side
Section 01
Why AdTAO MCP — and why it isn't free.
You can already paste exports into Claude, connect Google's official Ads MCP, or run a GAQL-wrapper MCP with dozens of tools. Each one stops at the same wall:
- →
Claude + a pasted CSV fast narrative — but no grounding, no peer context, and no safe way to act.
- →
Google's official Google Ads MCP real API reads of your accounts — read-only, no cross-advertiser corpus, no outcome history, no staged approval path.
- →
A GAQL-wrapper MCP lets the agent write queries and round-trip many times — faster automation, but still only the raw rows from accounts you already manage. Same blind spot.
None of those setups can answer:
- ·
"Is this a me problem or a market problem?"
- ·
"What actually happened when accounts like mine made this change?"
- ·
"What pattern is this account in, and where is it heading?"
— because those answers need years of pooled, privacy-governed advertiser data and a decision engine on top, not another GAQL pass.
|
Free / included elsewhere
|
What you pay AdTAO for
|
| Claude + pasted CSV | Composition and fluency — not peer data or verified benchmarks |
| Google Ads MCP | Your rows via GAQL; no cross-account pools; read-only; no outcome memory |
| GAQL-wrapper MCPs | More tools, same single-account ceiling; the agent does the query work |
| AdTAO MCP This product | Corpus-backed benchmarks, archetype/pattern intelligence, intervention history, composed server-side answers, human-gated writes |
Claude and Google give your agent access. AdTAO gives it judgment — the network data and outcome history required to manage a multi-account book without flying blind.
Section 02
What your agent gets that Google can't.
Decisions that require the network — not another spreadsheet paste.
01
See what your accounts can't see alone
Google's Ads MCP hands your agent your own rows — it has no peer benchmarks to give. Even Google's own UI only shows blunt, aggregated industry numbers, never a verified peer cohort you can interrogate by category, geo, and structure. AdTAO pools 20,000+ connected accounts, 1.3B+ search-term records, and 35M+ placement rows so your agent can answer market-vs-me questions, surface placement and query risk from network history, and spot competitive gaps — decisions you cannot make from a CSV of your own accounts.
Ask your agent things like
- →
"Did my CPA spike because the market moved or because we broke something?"
- →
"Which placements are wasting money in accounts like mine before they waste mine?"
- →
"Where are competitors winning queries I'll never see in my own search-term report yet?"
02
Your account as a picture, not a spreadsheet row
AdTAO classifies what your account looks like right now against hundreds of observable patterns drawn from 20,000+ advertisers — measurement health, structure, bidding state, spend shape, and more. Each pattern is a snapshot, not a verdict. Combined with your history and goals, your agent gets a ranked picture of what matters and what to do next — not a generic audit checklist.
03
Know what tends to happen before you approve the change
When you stage a change, AdTAO draws on real intervention outcomes from accounts in similar situations — not model guesswork. If we don't have enough history, we say so. And unlike read-only or paste-in setups, AdTAO can make the approved change for you — staged first, executed on your say-so — so you skip the manual round-trip back into the Ads UI. The commit is always yours.
Section 03
Proven — not promised.
Five real manager prompts. Four setups. Same model. You can run them yourself.
We don't ask you to trust a tagline. We run five adversarial scenarios that trip up every other setup we've tested — pasted CSVs, Google's official Ads MCP, and GAQL-heavy wrapper MCPs — then score every answer the same way: fabrication audit, independent judge (run twice), and safety on "make this change now" prompts.
Each scenario has a deliberate trap — hidden tracking faults, bulk changes on ineligible accounts, niche benchmarks that don't exist, learning-period campaigns. AdTAO is built to fail those traps differently: staged, approve-then-execute writes instead of blind execution or hand-cranked edits; insufficient_data instead of invented "industry ranges"; composed server-side answers instead of six GAQL round-trips.
Run date published. Methodology published. Caveats published. Our scores came from a controlled run on internal test accounts (we don't share them — you can't replicate our exact ground truth, and we won't publish those accounts). You rerun the same five prompts on your Google Ads book and judge safety, honesty, and speed yourself.
|
What we measured
|
AdTAO MCP
|
Others (typical)
|
| Outcome on "make changes" prompts | staged for your approval, then executed for you | advice only — no execution path, no approval gate |
| Refused to invent a benchmark with no data | yes (insufficient_data) | no — models invent confident ranges |
| Median time to answer | ~3.5 s | ~6 s bare Claude |
| Unsupported factual claims (lower is better) | fewest in the run | most: bare Claude |
Single run (2026-05-29); a 5-scenario smoke, not the full 36-scenario suite (in progress). One Google MCP leg ran without credentials, so its timings are degraded — we keep the full per-setup timings on the comparison page with that caveat attached rather than feature them here. Exact per-setup fabrication counts publish with the next run, alongside raw transcripts.
Challenge
Run the test. Challenge us.
Run our five published prompts on your accounts — AdTAO MCP vs whatever you use today. If AdTAO doesn't win on the scenarios where there's a clear winner, email mcp-challenge@adtao.io with the evidence. For any objective failure we can validate — wrong data, wrong Google Ads facts, or a provably worse recommendation — we'll apply account credit equal to one month on your current plan.
Connect
Hosted · streamable HTTP
Endpoint
mcp.adtao.io
Bearer-token auth via your API key. Walk-through: quickstart.
Verified hosts
- → Claude Code (CLI)
- → Claude Desktop
- → Cursor
Our principles
Principle
Human-in-the-loop
Every write is staged for your approval before anything hits the API.
Principle
Honest about limits
When the API can't do something, or the data is thin, we say so.
Principle
Grounded in the network
Answers use peer data and pattern intelligence; when data is thin, we say so instead of inventing benchmarks.