Windsor.ai vs Supermetrics for Claude: We Paid for Both, and Only One Answered

Measured on our own accounts between 26 July and 14 August 2026. Every claim below is labelled: tested means we ran it and have the output, documented means it comes from the vendor’s own pages or payloads, reasoned means we are inferring and saying so.

We sat down on 13 August to do something simple. Ask the same question, on the same Google Analytics property, over the same thirty days, through both Windsor.ai and Supermetrics, and print the two answers side by side.

Windsor answered in a second and a half: thirty rows, one per day, real numbers.

Supermetrics did not answer. It let us complete a brand new OAuth connection to the property — the confirmation screen said the connection had been successfully created — then refused to read it:

The AI chats destination is not included in your subscription 1864081 on team [our team].

We pay Supermetrics €59.78 a month. We pay Windsor €20.70 a month. On our subscription the €59.78 does not reach Claude at all — and when we went looking for why, the answer was more interesting than a missing feature. That is not the article we planned to write, and it is a better one.

What this comparison is, and what it is not

This is not a setup guide either — we already wrote those, one for connecting Windsor.ai to Claude and one for connecting Supermetrics to Claude, and the step-by-step lives there.

It is not a feature matrix. Both of these products are marketing data pipelines, both of them are competent, and both of them will move numbers from an ad platform into a place where you can use them. If you are an agency with fifteen clients, your criteria are not ours and this piece will not settle your decision.

This is a comparison run by one person, on four data sources, with a real card. We have paid both invoices. We are about to cancel one of them, and we will tell you which and why at the end.

One thing to declare before anything else, because it would be visible from a mile away if we buried it: we earn an affiliate commission on Windsor and nothing on Supermetrics. That is exactly the situation in which you should read a comparison slowly. So every number below is dated, every claim is labelled, and the section on where Supermetrics is genuinely better is not a courtesy paragraph at the bottom — it contains a finding that costs us money to publish.

Ignore the source count. It is the wrong number.

Every aggregator leads with how many platforms it connects to, and the figure is almost useless.

Here is what we read on 13 August 2026, from the products themselves rather than from a listicle. Supermetrics’ discovery call returns 172 sources to our account, while its own upgrade page, opened the same morning, offers “up to 158 data sources” on the top tier. Windsor’s tool manifest inside Claude declares 353, while its entry in Claude’s connector directory describes four named platforms “+ 320 more”. Documented, 13 August 2026. Both vendors disagree with themselves across two of their own surfaces on the same day, and we did not count either catalogue by hand.

None of those four numbers changed our decision by a single euro, because what matters is not how many sources exist. It is how many you are allowed to connect at the price you are paying. On that question the two plans are closer than the catalogues suggest: Windsor Basic allows three data sources. Supermetrics Starter also allows three, with three accounts per source and one user. We had four sources and both plans forced us to drop one. The catalogue size was never the constraint; the seat at the table was.

Worth saying plainly: there are more than two of these in Claude’s directory. Funnel is in there too, describing itself as cross-channel analytics across 600+ connectors, and we have never tested it. We are comparing the two we have paid for.

When both of them work, they agree to the cent

On 26 July 2026 we ran the same question through both systems against the same Meta ad account, for the campaign period of 7–12 July. Three metrics: impressions, spend, link clicks.

3,365 impressions. €35.61 spent. 292 link clicks. Identical on both, row by row. Tested, 26 July 2026.

If that €35.61 looks familiar, it is the same campaign we wrote about in our count of Claude’s marketing connectors — a small boosted post we thought had been switched off months earlier, and which turned out to be the cheapest test account we could have asked for.

That is the finding that frames everything else. There is no quality gap in the numbers. Anyone telling you that the expensive pipeline returns better data than the cheap one is describing something other than what we measured. Both are reading the same platform APIs, and the platform does not know which reseller is asking.

If you want the honest caveat, here it is: that measurement is now impossible to repeat, from both directions. Windsor no longer has Facebook Ads connected, because Basic holds three sources and we chose the other three. Supermetrics still has Facebook Ads connected and will not serve the data. The window closed while we were standing in it.

Same question, different meaning: a trap in the field names

Before you compare two data tools, check that the field you are comparing means the same thing in both. This one caught us on 13 August and it is genuinely subtle.

Windsor’s GA4 connector describes source as “the source attributed to the session“. Supermetrics’ GA4 connector calls the field of the same name “Event source” and describes it as “the source attributed to the conversion” — session scope lives in a separate field called sessionSource. Documented from both connectors’ own field descriptions, 13 August 2026.

So the naive comparison — source against source — puts session-scoped traffic next to conversion-scoped traffic and calls the difference a discrepancy between vendors. It isn’t. It is a difference between two definitions that share a label.

This is the practical reason to distrust any side-by-side you read online, including ours if we ever stop showing our parameters. Two systems returning different numbers is not evidence that one is wrong. It is usually evidence that somebody asked two different questions.

We have been caught by this ourselves, and the receipt is worth showing. Windsor once returned one campaign out of thirty-eight on a Meta account, and we read it as a defect in Windsor. It was not. When we re-asked both systems the identical question — same window, same fields, same account, which our first attempt had not done — the missing campaigns turned out to be an artefact of Meta’s reporting endpoint, which answers only for windows in which a campaign actually delivered. The finding survived; the blame moved. Tested, 6 August 2026.

What actually happened when we asked

Our Supermetrics setup has two separate things in it, and until 13 August we thought we understood how they related. There is a Starter subscription at €59.78 a month including VAT. And there is a licence the connector reports as DEMO_CLAUDE, which ran from 24 July to 7 August and now reports INVALID with zero days remaining.

What we assumed, and wrote down as a reasoned guess: the Starter and the Claude connector are separate products, so the Starter probably does not enable the connector. What we could then prove, because we finally had an authenticated source to point it at:

• Metadata calls still work with the licence expired. Listing sources, listing accounts, listing fields — all fine.
• New onboarding still works. The connector handed us a fresh login link and completed a full OAuth connection to a Google Analytics property. Six days after the licence expired.
• Data calls do not work. Every one returns the subscription error quoted at the top of this article.

Tested, 13 August 2026. We checked it was not specific to the new connection by running the same call against Facebook Ads, which has been authenticated for weeks and returned real data on 6 August. Identical error. The gate is on the subscription, not on the source.

Two things follow from that, and the second one matters more than the first.

The small one: our reasoned guess was right in outcome. The Starter subscription we pay for does not reach Claude.

The large one is that the reason is not the one the error implies, and it is not a hidden fee either. We opened the subscription page. Supermetrics Starter is €49 a month, which becomes €59.78 on our invoice with Italian VAT at 22% — the two numbers reconcile exactly. The plan includes three data sources, three accounts per source, one user, and one core destination of your choice. And the list of core destinations you can choose from includes Supermetrics Studio, Looker Studio, Google Sheets, Excel, Power BI — and “Claude, ChatGPT & more.”

Ours is Supermetrics Studio. That is the whole story. The connector is not missing from the plan; it is one of the doors we did not pick, and a fourteen-day demo licence kept it open long enough that we never noticed we had chosen a different one. Documented from the vendor’s own subscription and upgrade pages, 13 August 2026.

If you want both — a Studio dashboard and Claude access — the price is on the same page, in a table called Extras: a second destination on Starter costs +€49 a month. So the honest like-for-like against Windsor Basic at €20.70 is not one number but three, depending on what you need: €49 plus VAT for Claude access alone, €98 plus VAT for Claude access alongside a dashboard destination, and €20.70 for the Windsor plan that carries three sources and the Claude connector together with no destination to choose.

One thing we did not test, and will not guess at: whether you can switch your core destination mid-subscription, and at what cost. There is a visible control for swapping data sources. We found no confirmed equivalent for destinations, and we were not going to experiment on a live subscription a week before cancelling it — swapping away from Studio would have destroyed the destination we were about to measure. If you are choosing today, choose the destination carefully, because we cannot tell you how reversible it is.

There is a third thing, smaller but worth noticing if you are the sort of person who keeps track of what a system calls itself. Across one account in a single morning we were shown three different identifiers for what is nominally one relationship: a licence number, a product code, and a subscription number that appears only inside an error message. We have seen the same shape at the other vendor, where a single account state was called trial by a popup, Free by a badge and FREE by the API on the same screen. Neither of these is a bug. Both make it harder than it should be to answer the question “what am I actually paying for.”

Two ways of going wrong

Any tool works when everything is in order. The comparison that decides whether you can trust one is what it does when it cannot answer you — and in the space of one week we have watched both of these fail on our own accounts.

Supermetrics tells you. Every response it returns, including the successful ones, carries an upsell string in a notes field: a line saying the trial has expired with a link to pricing. When it cannot serve data, it says so in an explicit error naming the subscription. It is noisy, it is commercial, and it is honest. You cannot mistake a refusal for an answer. Tested, 9, 13 and 14 August 2026.

Windsor does not. On 10 August our trial ended overnight and the account dropped to the free plan, which allows one data source against the four we had connected. Nothing disconnected. Nothing errored. Every query succeeded. What changed was the content:

• Numeric fields came back as zero.
• The date range was replaced rather than emptied — a 30-day preset, a two-year preset and explicit start and end dates each returned one row, dated today.
• Text fields contained an advertisement instead of a value. Where a traffic source should have been: “Uh-oh! You’ve connected more accounts than your Free plan allows. Upgrade here…”

Tested, 10 August 2026. At exactly one source, the identical query returned thirty days of correct, properly labelled data. The free plan works perfectly inside its fence and does not tell you when you have stepped outside it.

Sit with what that means for the specific case of an AI assistant reading the payload. It does not see a limit. It sees data. Ask it where your traffic came from and it will tell you, fluently, that your top source is called “Uh-oh! You’ve connected more accounts…” and sent you zero visitors.

An upsell you can see is an annoyance. An upsell in the value field is a wrong answer with good grammar. If we had to pick one axis on which one of these products is clearly better designed than the other, it would be this one — and it is the one where the product we do not earn from wins.

Write access: one lever versus none

Both of these can do more than read, and the controls are very different.

Windsor has a single toggle in the team settings called “Enable write actions via API.” It is off by default. When it is on, it is on for everything — twenty-two write actions across Meta advertising and Instagram, including publishing feed posts, videos, carousels and stories to your own Instagram profile, and commenting as the account. No granularity, one lever, and you can see it. That count was twenty when we first checked on 6 August. The two Instagram additions appeared some time between those two checks, and nothing on our side changed. A single lever is easy to understand, and what sits behind it grows without asking you again. Tested, 14 August 2026; ours is off.

Supermetrics has no such toggle that we could find. Its discovery payload marks six advertising platforms — Google Ads, Microsoft Advertising, Facebook Ads, LinkedIn Ads, TikTok Ads and ChatGPT Ads — with campaign_management_supported: true, and its own notes state that those platforms support campaign creation and updates through dedicated tools. That was returned to us on an account with an expired licence. Documented from the connector’s payload, 13 August 2026.

We did not test whether those write tools actually execute, and we want to be explicit about why: the only way to test a campaign write is to write to a real campaign in a real ad account. We are not doing that to make an article. If you connect this, find out where the controls are before you find out whether they work.

Where Supermetrics is genuinely better

Three places, and the first is the real one.

Destinations — and the proof arrived sideways. On 14 August, in one sitting, the same subscription did two apparently contradictory things. It refused a data query into the chat, with the subscription error at the top of this article. Then it accepted an instruction to build a hosted dashboard — a live board on Supermetrics Studio, reading the same Google Analytics property that had just been refused — and served it. We opened the link and got thirty bars, one per day, real numbers, re-queried on every view. Tested, 14 August 2026.

That is the clearest statement of what this product is that we have found anywhere, and it is not on a pricing page. Nothing is broken and the expired licence is not the point. The subscription pays for a destination, and ours is not the conversation: data that will not travel into a chat travels perfectly into the artefact we bought.

Here is the correction we owe Windsor in the same breath, because we assumed this category was empty on its side and it is not. Windsor’s Claude connector exposes fourteen destinations — Google Sheets, Airtable, BigQuery, Snowflake, PostgreSQL, MySQL, Supabase, Azure SQL, Redshift, Databricks, Amazon S3, Azure Blob and two of its own data stores — each with its own field schema. We opened one of them in full, Google Sheets, and it offers schedules down to every fifteen minutes. What none of them will do is finish the job from the chat: all fourteen come back flagged as not creatable in conversation, and the transfer has to be set up in the dashboard first. And one name is absent from that list altogether, which is Looker Studio. Tested, 14 August 2026.

So the shape of this advantage is narrower and much firmer than “Supermetrics does destinations.” Both of them do. If your deliverable is a spreadsheet or a warehouse table, that is not the reason to pay three times as much. If your deliverable is a Looker Studio report that a client opens on a Monday, only one of these two arrives, and price stops being the deciding factor.

It fails out loud. This has its own section above, and it belongs in this column too: a refusal you can read beats a zero you cannot.

Scale and history. Agency volumes, many accounts, long backfills. We cannot test this and we are not going to pretend to. Reasoned.

And one honest deduction from that column, because balance is not the same as flattery. Supermetrics Studio hosts dashboards you can build inside a chat, which sounds like the answer to “send the client a live link.” We built one on 14 August and opened the sharing dialog. General access offers three settings: restricted, everyone on your team, and anyone with the link — and the third is greyed out, behind a padlock, labelled coming soon. So on the day we looked, sharing a board meant inviting named people one at a time or opening it to your own team. The public link is on the roadmap, not in the product. Tested, 14 August 2026, and consistent with what we found on 26 July. That is worth knowing before you build a client workflow on it.

So which one

Disclosure: Some links in this guide are affiliate links. If you start a plan through one of them, we may earn a commission at no extra cost to you. No vendor sees or approves our guides before they go live.

If you are one person with a handful of sources and you want your own numbers inside Claude, Windsor is the cheaper answer by a wide margin and it is the one we kept. The trap to plan for is not the price, it is day thirty-one: write the trial expiry somewhere outside the tool, and decide in advance which single source you would keep, because coming back down is easier than going back up. We wrote up that pattern, and the three other ways a free connector ends up costing money, in what MCP connectors actually cost.

If your output is a scheduled Looker Studio report, Supermetrics is the one of the two that gets there, and the comparison stops being about the connector at all. If your output is a warehouse table or a spreadsheet, look at Windsor’s destination list before you assume the expensive product is the only route. And either way, be deliberate about the destination you pick, because on the entry plan you get one — and if Claude is meant to be part of the workflow, that is a decision you make at checkout rather than a switch you flip later.

And if you have exactly one data source, look hard at whether you need either. Windsor’s permanent free plan holds one source and works correctly inside that limit. That is a real option, and neither vendor’s pricing page is going to lead with it.

What we did with our own money

On 20 August we are cancelling the Supermetrics Starter subscription, and we are keeping Windsor Basic at €20.70 a month, billed monthly rather than annually because a business with no revenue should not be buying twelve months of certainty.

The reason is narrow and we want to state it precisely rather than generously. We are not cancelling because Supermetrics is a worse product, and we are not cancelling because it hid a fee — it did not. We are cancelling because the one destination our plan carries is a dashboard editor we stopped using — one that works, as we confirmed on 14 August, and that we do not need — while the destination we actually need would have cost the same again, and a product that does the job we have costs €20.70.

The subscription page confirms access runs to 22 August whether we cancel on the 20th or not, which is the only reason we can afford to look around first.

We will update this article if that changes, with the date on the line.

You Might Also Like

If this was useful, these go deeper on the pieces we referenced above.