Run your ads from ChatGPT or Claude.
Connect your ad accounts. Ask questions, build reports and make changes from one conversation, across every platform at once.
- Free forever, no card
- 1 ad account, 200 tool calls
- Read only until you upgrade
What changed across my accounts this week?
Spend
18,240EUR
+12%
Orders
612
−8% against you
CPA
29.80EUR
+22% against you
Main drivers
- Halcyon Prospecting: CPA up 37%, frequency past 4.
- Northwind PMax absorbed 41% of the budget after Tuesday.
- GA4 recorded 8% fewer checkouts than the platforms claimed.
Pause the underperforming ad set.
Change ready to applynot applied
tiktok_update_adgroup_status“Preview only, nothing was changed.” Repeat the call with confirm: true to apply it.
What you can do
Everything you need,in one conversation.
Analyse performance, work out what went wrong, write the report and make the change, without opening six dashboards or exporting another spreadsheet.
Six platforms in. Your assistant out.
- Meta Ads
- Pinterest Ads
- Google Ads
- TikTok Ads
- Search Console
- GA4
Get MCP Ads
mcp-review.getmcpads.com/mcp
- Claude
- ChatGPT
- Cursor
- Codex
6 platforms, 236 tools, one address.Your assistant asks. We answer for all of them.
Ask anything about your performance, and get the reason behind the answer.
ExampleWhy did our CPA jump last week?
CPA, last 8 weeks
+32% against youprevious level
Creative fatigue on two prospecting ad sets. Frequency crossed 4 on 12 March and CTR fell against their own launch baseline.
Turn the data into a report you can send, in the time it takes to ask.
ExampleBuild the weekly summary for Northwind.
Weekly performance summaryeur
Spend
24,180
+9%
Revenue
96,700
+14% in your favour
ROAS
4.00
+5% in your favour
daily spend
rule: last week’s average
Ask for it as a table, a summary, or a paragraph you can paste into the client thread.
Make changes with a preview first, and an approval after it.
Halve the budget on the losing ad set.
Previewmeta_update_adset_budget
- 01The first call describes the change and stops. Nothing has moved.
- 02The second call carries confirm: true, and only then does the platform hear about it.
One question. Every platform, in the same answer.
ExampleSpend and revenue by platform, this month against last.
Nothing is green because it went up. A filled figure is one that moved against you; spend rising is neither good nor bad until you know what it bought.
201 media-buying playbooks written for the agent, not for you.
- 288
- decision rules
- 263
- failure modes
- 222
- hand-backs
Your accounts stay yours.
- Read-only by defaultWrites are off until you turn them on, per workspace.
- Per account, not per connectionAn MCC or a Business Manager opens nothing underneath it.
- Every call is recordedTool, account, and what came back. Yours to read.
- Your assistant never holds an ad tokenCredentials are encrypted at rest and never leave the server.
Nothing changes until you say so, twice.
Nothing changes until you approve it
Every write previews first, then stops.
Preview every action
The exact target, the exact new value, before the call.
You stay in control
Turn writes off for a workspace and the tools disappear.
Connecting is free. One ad account, no card.39 € a month opens the other 3 and the writes.See what is in each plan
Every tool name, every skill name and every count above is read off what ships. The three panels marked Example use invented accounts and invented figures. The gateway has no demo data, and no number on this page is a claim about what a campaign will return.
Platform coverage
Six sources. Everything anaccount manager opens every morning.
Paid and organic in the same conversation, so the assistant can put spend next to the traffic it bought without you exporting either.
236
tools live today
65
of them can write
- Meta Ads57 tools
Facebook and Instagram delivery, creatives, audiences, lift studies, catalogue joins.
23 write
- Pinterest Ads32 tools
Delivery and catalogue reporting, targeting analytics, audience insights, billing.
5 write
- Google Ads44 tools
Raw GAQL, search terms, Shopping, Performance Max diagnostics, change history.
10 write
- TikTok Ads58 tools
Auction reporting, creative fatigue recipes, audience overlap, Spark Ads joins.
27 write
- Search Console18 tools
Search Analytics, cannibalisation, query clustering, indexation watchlists, sitemaps.
read only
- Google Analytics 427 tools
Reports, realtime, funnels, key events, ecommerce and server-side tagging diagnostics.
read only
- Microsoft Advertising
OAuth is in place. The tool set is not shipped yet, so nothing is exposed to your assistant.
not shipped
Skill library
Tools are access.Skills are the job.
Any gateway can hand an assistant a list of endpoints. Which one to call, in what order, at what number a conclusion still holds, and when to stop and ask you — that is the part nobody ships. It is written down here, two hundred times.
201
skills written
26
deep contracts
Three of them, exactly as written
Meta Ads
- When
- CTR is down 30% or more against the ad's own first-14-days baseline AND frequency exceeds 4 on a 7-day window
- Conclude
- Fatigue confirmed. Rank by current spend.
- Unless
- CPM rose sharply at the same time, which can depress CTR account-wide; check a non-fatigued control ad first.
Creative fatigue4 decision rules
Google Analytics 4
- When
- A key event used for bidding fired zero times in 7 days on a site with traffic
- Conclude
- Broken wiring, highest severity: platforms are optimising toward silence.
- Unless
- No exception — this one holds outright.
Tracking audit4 decision rules
Google Ads
- When
- Waste candidates sum to under 2% of search spend
- Conclude
- Report the account as clean; do not pad the list to look useful.
- Unless
- No exception — this one holds outright.
Search waste3 decision rules
- 288quantified decision rules
- 263named ways it goes wrong quietly
- 222lines that hand the decision back
201 of them, by what they need
Every skill sits under the source its work cannot happen without, so these seven add up to the figure in the heading rather than past it.
31
The last group
Needs no single account. Those read whatever you have connected and reconcile it.
Your assistant finds them on its own
Nobody has to browse this library. Every MCP endpoint announces search_skills and get_skill, so the assistant finds the contract that fits the question and loads it whole — procedure, thresholds, failure modes, and the line where it stops and asks you.
What you can ask
One question, five calls,one table back.
No export, no pivot, no window switching. This is a whole turn — what you type, what the gateway does about it, and what lands in the conversation.
https://mcp-review.getmcpads.com/mcpYou
Spend and CPA by campaign across every platform, last 7 days against the 7 before.
The assistant
It reads the catalogue, plans five calls and makes them in the same turn. Nothing was exported, and nothing was configured beforehand.
- 01
list_accountsFirst, always. It says which accounts are reachable before any platform is called. 3 accounts - 02
google_ads_get_insightsCampaign performance over both windows, in the account currency. 18 campaigns - 03
meta_get_insightsThe same on Meta, planned against Meta's own breakdowns. 11 campaigns - 04
tiktok_get_insightsThe same on TikTok. 6 campaigns - 05
pinterest_run_reportRouted through async reporting, because the window is wider than the sync endpoint takes. 4 campaigns
What comes back
- Northwind — Search Brand4,120 EURCPA 11.40−6%
- Northwind — PMax Retail9,860 EURCPA 28.10+19%
- Northwind — Advantage+6,240 EURCPA 24.70+4%
- Halcyon — Prospecting3,180 EURCPA 41.90+37%
- Halcyon — Catalogue1,450 EURCPA 19.20−11%
| Campaign | Spend | CPA | vs prev |
|---|---|---|---|
| Northwind — Search Brand | 4,120 | 11.40 | −6% |
| Northwind — PMax Retail | 9,860 | 28.10 | +19% |
| Northwind — Advantage+ | 6,240 | 24.70 | +4% |
| Halcyon — Prospecting | 3,180 | 41.90 | +37% |
| Halcyon — Catalogue | 1,450 | 19.20 | −11% |
Metric names differ on every platform and are not silently unified. The table lines up what is comparable and says so when something is not.
You, ten seconds later
Why is Halcyon's TikTok CPA up? Show me the ad sets.
Two more calls, on the same data, in the same conversation. tiktok_get_insights at ad-set grain and tiktok_get_creative_fatigue_recipes for what the platform's own diagnostics say to do about it. This is the turn a spreadsheet cannot take: the follow-up costs nothing, so you ask the second question.
/tools. The accounts, the campaign names and every figure in the table are invented for this page — the gateway has no demo data, and nothing here is a claim about what a campaign will return.22 questions
Grouped by the job rather than by the platform, because nobody wakes up wanting to use Meta's API. Every tool named on a card ships today and has its own page.
The questions that normally cost an afternoon of exports and a pivot table.
All three, on one pageWhich TikTok creatives are fatiguing, and what should we swap them for?
Reads ad-level daily reporting, spots the frequency and CTR decay pattern, and returns the refresh recipes TikTok's own diagnostics imply.
tiktok_get_creative_fatigue_recipestiktok_get_insightstiktok_get_creativesShow me search terms we paid for last month that we would never have bid on.
Pulls the search term report, joins it to keyword performance, and lists spend against terms with no conversion and no match to your keyword set.
google_ads_get_search_termsgoogle_ads_get_keyword_performancegoogle_ads_get_budgetsIs any Performance Max campaign spending without the ad strength to justify it?
Reads asset group diagnostics, ad strength, coverage action items and placement types, then ranks campaigns by spend over asset health.
google_ads_get_pmax_asset_diagnosticsgoogle_ads_get_pmax_placementsgoogle_ads_get_pmax_assetsWhich of our pages are competing against each other in Search?
Runs the query-page cannibalisation detector over Search Analytics and clusters what is left by brand, question and category intent.
gsc_detect_cannibalizationgsc_cluster_queriesgsc_query_search_analyticsAre two of our TikTok ad groups bidding against each other?
Compares targeting fields, custom audiences and saved audiences across ad groups and reports where the overlap is large enough to matter.
tiktok_get_audience_overlaptiktok_get_adgroupstiktok_get_audience_detailsWhy has this Meta ad set stopped spending its budget?
Reads Meta's own delivery diagnostics for the ad set, then the audience size, the schedule and the recent insights, and names the constraint rather than guessing at it.
meta_get_delivery_diagnosticsmeta_get_adsetsmeta_get_insightsDid our brand lift study actually prove anything?
Pulls the study, its results and the confidence intervals, then reads them back in plain language, including when the answer is that the test was underpowered.
meta_get_ad_studiesmeta_get_study_resultsmeta_interpret_experiment_resultsIs our GA4 property measuring what we think it is?
Runs the ecommerce, server-side tagging and key event diagnostics in one pass, and lists the events that are declared but never fire, or fire without the parameters the reports expect.
ga4_get_ecommerce_diagnosticsga4_get_server_side_tagging_diagnosticsga4_get_key_eventsWho changed what in this Google Ads account in the last two weeks?
Reads the change event log inside the 30-day window Google allows, with the actor, the resource and the field that moved.
google_ads_get_change_eventsgoogle_ads_get_campaignsCross-platform intelligence
One question.Every platform.
Six sources in one reasoning context is not six integrations in a list. It is the difference between asking what Meta reported and asking whether anyone else saw the same thing.
- Ask naturallyPlain English, in the assistant you already pay for. No query language, no export, no tab you forgot to refresh.
- All your sources at oncePaid and organic in the same reasoning context, so spend can sit beside the traffic it bought without either being pasted anywhere.
- One clear answerOne table, with the platforms named and the limits stated. Where two sources disagree the answer says so rather than picking one.
Which platform actually returned money last month?
Revenue is each platform’s own reported figure. Attribution windows differ and are not reconciled behind your back.
- Meta Ads
- Pinterest Ads
- Google Ads
- TikTok Ads
- Search Console
- Google Analytics 4
Read live from each platform’s own API.
The work it replaces
The Monday morning numbers, pulled from four advertising APIs and two analytics APIs in the same turn.
Before Get MCP Ads
- 6 dashboards
- 4 exports
- 1 spreadsheet
- Hours a week
With Get MCP Ads
- 1 question
- 1 conversation
- 1 answer
- Minutes to act
Safe Writes
Nothing changes untilyou say it twice.
An assistant composing an API call can pick the wrong campaign, or the wrong order of magnitude on a budget. So no write tool applies anything on the first call. It describes the change it would make, and stops.
confirm: false{
"applied": false,
"action": "meta_update_adset_status",
"change": {
"target": "23861049518",
"newStatus": "PAUSED"
},
"message": "Preview only, nothing was changed.
Repeat the same call with confirm: true to
apply this change to the live account."
}confirm: true{
"applied": true,
"action": "meta_update_adset_status",
"result": {
"success": true
}
}result is the platform's own response, passed through untouched.01
Every write previews first
Called without confirm, a write tool returns the change it would make — the account, the object, the old value and the new one — and stops.
02
Applying takes a second, explicit call
The same tool, the same arguments, plus confirm: true. That second call is the moment a human reads what is about to happen, which the protocol guarantees nowhere.
03
Campaigns are always created paused
On all four ad platforms, and there is no parameter that changes it. Something you did not read cannot start spending while you are reading something else.
04
The Free plan has no writes at all
Not previewed writes, not throttled writes — none. Trying the product cannot move a budget, because the tools that would are not in its catalogue.
Why Get MCP Ads
Why not just use theofficial platform MCPs?
Where an official server exists it is the reference for its own API, and the comparison is structural rather than qualitative: it answers for one platform, and the questions media buyers ask span several.
One endpoint for every platform. Advantage: Get MCP Ads.
- Get MCP Ads
- One URL, or one per source if you prefer
- The official platform MCP servers
- One server per platform, each with its own auth
Cross-platform questions in one context. Advantage: Get MCP Ads.
- Get MCP Ads
- One reasoning context across 6 sources
- The official platform MCP servers
- Requires wiring several servers and reconciling their conventions
Shared account scoping. Advantage: Get MCP Ads.
- Get MCP Ads
- One dashboard decides what every tool can reach, enforced server side
- The official platform MCP servers
- Per server, when offered
Consistent tool semantics. Advantage: Get MCP Ads.
- Get MCP Ads
- Same conventions and the same preview-then-confirm writes everywhere
- The official platform MCP servers
- Each server names and shapes things its own way
Analytics next to the ad data. Advantage: Get MCP Ads.
- Get MCP Ads
- Both are full sources among the six
- The official platform MCP servers
- GA4 and Search Console are separate servers again
Availability. Advantage: Get MCP Ads.
- Get MCP Ads
- All six sources live
- The official platform MCP servers
- Some are beta, pilot, or capped
Depth on its own platform. Advantage: The official platform MCP servers.
- Get MCP Ads
- 236 tools built on the public APIs
- The official platform MCP servers
- The vendor's own surface, first to get new API features
Cost. Advantage: The official platform MCP servers.
- Get MCP Ads
- Free plan, then from 39 EUR a month
- The official platform MCP servers
- Generally free
Third party in the token path. Advantage: The official platform MCP servers.
- Get MCP Ads
- The gateway, unless you self-host the adapters
- The official platform MCP servers
- None
| Capability | Get MCP Ads | The official platform MCP servers |
|---|---|---|
| One endpoint for every platform. Advantage: Get MCP Ads. | One URL, or one per source if you prefer | One server per platform, each with its own auth |
| Cross-platform questions in one context. Advantage: Get MCP Ads. | One reasoning context across 6 sources | Requires wiring several servers and reconciling their conventions |
| Shared account scoping. Advantage: Get MCP Ads. | One dashboard decides what every tool can reach, enforced server side | Per server, when offered |
| Consistent tool semantics. Advantage: Get MCP Ads. | Same conventions and the same preview-then-confirm writes everywhere | Each server names and shapes things its own way |
| Analytics next to the ad data. Advantage: Get MCP Ads. | Both are full sources among the six | GA4 and Search Console are separate servers again |
| Availability. Advantage: Get MCP Ads. | All six sources live | Some are beta, pilot, or capped |
| Depth on its own platform. Advantage: The official platform MCP servers. | 236 tools built on the public APIs | The vendor's own surface, first to get new API features |
| Cost. Advantage: The official platform MCP servers. | Free plan, then from 39 EUR a month | Generally free |
| Third party in the token path. Advantage: The official platform MCP servers. | The gateway, unless you self-host the adapters | None |
The dot marks the column that takes the row. 6 of them are ours and 3 are theirs, which is first-party depth on a platform, cost, and the fact that no third party holds the token. Read on 2026-08-31 against mcp.facebook.com/ads, github.com/modelcontextprotocol/servers
Stay with the official servers if
- You work on exactly one platform. A gateway adds a layer and a bill for nothing.
- You want the vendor's own surface, including whatever they ship next week.
- You would rather no third party ever held a token for your accounts.
- Cost decides. Official servers are generally free.
Look past one platform if
- Your questions cross platforms
- You manage accounts for other people
- Availability is gated
- Write behaviour differs per server
Proof
What is true here, andhow you can check it.
No customer count, no logo wall, no quote from someone who does not exist. Four guarantees the gateway enforces on every call, and six documents you can open right now and hold this page to.
01
Your assistant never holds an ad token
It talks to the gateway; the gateway talks to Google, Meta, TikTok and Pinterest. Ad platform credentials are encrypted at rest with AES-256-GCM and never leave the server, so a leaked chat transcript is not a leaked ad account.
02
Exposure is per account, not per connection
A Google MCC or a Meta Business Manager can sit over dozens of client accounts. Connecting one does not expose them all. The gateway answers only for the account IDs you explicitly ticked, and returns nothing for the rest.
03
Safe Writes: every change previews first
Every write tool returns a description of the change and stops. Applying it takes a second call with confirm set to true. An assistant that picked the wrong campaign shows you the wrong campaign instead of pausing it. Campaigns are always created paused.
04
OAuth 2.1, not an API key in a config file
The client registers itself dynamically, PKCE is required, and the code is bound to the redirect URI it was issued for. Revoking an assistant's access does not mean rotating a secret across your team.
Six things you can open
- the catalogue
Every tool, with its parameters
All 236 of them, generated from the gateway that is running rather than written for a page. If a tool is not there, this site does not claim it.
/tools - the protocol
OAuth 2.1, PKCE, and no API key
The full handshake, in the strings the gateway returns: the refusal, the discovery documents, dynamic client registration and the token exchange.
/docs/authentication - the scoping rule
What a tool refuses, and why
How an account becomes reachable, what happens to a call that names one you did not tick, and why a manager account is not a permission slip.
/docs/concepts/accounts - the write rule
Preview, confirm, and the shape of both
The two calls written out per platform, including the fields each preview returns and the errors you get for confirming something that has moved.
/docs/concepts/writes - for agents
The whole site, in Markdown
An llms.txt index, and a Markdown twin of every page here — so an assistant evaluating this product reads the same claims you do, not a scrape of them.
/llms.txt - the publisher
A named company, not a handle
Published by BENLY HLG, with terms, a privacy policy that says where the data sits, and a contact address answered by a person.
/about
Six repositories you can read
Every platform adapter is an open-source MCP server under Apache 2.0. Inspect it, run it on your own machine with your own credentials, or fork it. The hosted gateway runs the same tools, with the connection work done for you.
- Apache 2.0
Meta Ads
34 read tools and 23 write tools that preview before they apply.
@getmcpads/meta-ads-mcp-server - Apache 2.0
Pinterest Ads
27 read tools and 5 write tools that preview before they apply.
@getmcpads/pinterest-ads-mcp-server - Apache 2.0
Google Ads
34 read tools and 10 write tools that preview before they apply.
@getmcpads/google-ads-mcp-server - Apache 2.0
TikTok Ads
31 read tools and 27 write tools that preview before they apply.
@getmcpads/tiktok-ads-mcp-server - Apache 2.0
Google Search Console
18 read-only tools. This one cannot write at all.
@getmcpads/google-search-console-mcp-server - Apache 2.0
Google Analytics 4
27 read-only tools. This one cannot write at all.
@getmcpads/google-analytics-mcp-server
Pricing
Priced on what you connect,not on what you ask.
Every plan reaches all 6 sources. What grows is the number of ad accounts and seats, and whether the gateway is allowed to change anything.
Free
0no card
One account, read only, no card. Enough to see whether the answers hold up.
- 1 ad account
- 1 seat
- 200 tool calls per period
- Read only
Pro
39EURper month
One person, a handful of accounts, pausing and capping from the chat.
- 4 ad accounts
- 1 seat
- Unlimited tool calls
- Writes, with a preview before every change
Team
99EURper month
A desk that shares a portfolio: several people, a dozen accounts.
- 12 ad accounts
- 5 seats
- Unlimited tool calls
- Writes, with a preview before every change
Agency
Most teams199EURper month
A full agency, with the account and seat count that implies, and a support queue.
- 30 ad accounts
- 15 seats
- Unlimited tool calls
- Writes, with a preview before every change
- Priority support and tool requests
Free needs no card; every paid plan opens with 7 days free, and a year costs 10 months.
Questions
The ones that come upbefore the first call.
How it works
Connect once.Then just ask.
OAuth on this site, the accounts you tick, one address pasted into the assistant you already use. There is no config file and no key to distribute.
- 01
Connect a platform
OAuth on this site, not in your assistant. The refresh tokens are encrypted with AES-256-GCM before they touch the database, and the key is not stored beside them.
Google, Meta, TikTok, Pinterest- + more
- 02
Tick the accounts you want reachable
Connecting an MCC or a Business Manager does not open the client accounts underneath it. You choose them one by one, and un-ticking one removes the access rather than hiding the row.
Per account, per platformreachable accountsExample
- Northwind Retail483 992 1077
- Halcyon Studio771 204 5518
- Vireo Group902 663 4410
- 03
Paste one URL
Into Claude, Cursor, ChatGPT or Codex. The client registers itself over OAuth 2.1 on first use, so there is no key to generate, rotate or leak into a config file.
mcp.getmcpads.com/mcpadd custom connectorReconstruction
Remote MCP server URL
https://mcp-review.getmcpads.com/mcp
Authorised once. It stays until you revoke it.
https://mcp-review.getmcpads.com/mcpWhere it goes
- ClaudeSettings, then Connectors, then Add custom connector
- Claude CodeOne command, in any project or for every project
- ChatGPTSettings, then Connectors, with developer mode on
- CodexOne command, or four lines of TOML
Cursor, and anything else that speaks MCP over streamable HTTP with OAuth 2.1, registers itself on first use. There is nothing to install and no key to distribute — the quickstart has the exact clicks for each one.
Secure by design
OAuth 2.1 with PKCE. Refresh tokens are encrypted before they touch the database.
You stay in control
Read-only until you open the writes, and every write describes itself before it runs.
No new tool to learn
It runs inside the assistant you already use. There is no dashboard of ours to open.
Set up in minutes
One address, pasted once. The client registers itself on first use.
Connect one account,ask one question,then decide.
The Free plan is not a countdown. It is read-only access to all six sources, 200 calls per period, for as long as you want to keep testing the answers against what you already know.