Ask ChatGPT about your reviews
One URL, pasted once, and the question you were never going to build a dashboard for takes a sentence. ChatGPT queries your reviews live and shows you the ones it counted. Read-only, so it can analyse and can never post as you.
Set up in under five minutes • No demo required
ChatGPT Reviewflowz connector
Read only
What are people saying about onboarding since March?
keywords: ["onboarding", "setup", "import"]
start_date: "2026-03-01" · limit: 500
412 reviews mention it, averaging 3.42. Three complaints, in this order.
Priya N. · 11 May “The new wizard imported four years of invoices twice. We spent a week deleting duplicates before we could bill anyone.”
The question you never bothered to ask
Most questions about your reviews never get asked, and it is not because nobody wondered. “What are people saying about onboarding since March?” is a query somebody has to build, a pivot somebody has to maintain, and a dashboard nobody opens twice.
So the question gets dropped. Not weighed up and rejected, dropped: it was never worth half a day on a hunch. The hunches that would have been right die alongside the ones that would not, and you never learn which was which.
Connected to ChatGPT it costs a sentence. The answer comes back with counts you can check and the reviews that produced them, and it took long enough to read the reply and no longer than that.
Which changes what is worth asking. The reviews that would once have had to justify a whole reporting project are the ones you can now just ask about, on a Tuesday, because something occurred to you.
Are the one-stars about the same thing they were last year?
Two exports, a keyword pass, most of a Thursday
Which complaint did we quietly fix without anyone noticing?
A ticket, a queue, and an answer some time next week
Do the reviews we never replied to read differently?
A pivot table somebody has to keep alive
One sentence each, now.
To justify starting, you had to already suspect the answer. That is the wrong way round, and it is why the list above is empty.
What developer mode actually requires
Custom MCP connectors sit behind developer mode in ChatGPT, and developer mode needs a paid plan. The free plan cannot add one at all. Better to read that here than to discover it three steps in, once you have already found the server URL and named the connector.
On a Business or Enterprise workspace an admin turns it on before anyone else can, because it is a workspace permission rather than a personal one. Nothing shows up in your own settings until they do, which is worth knowing before you go hunting for a toggle that is not there yet.
Then you create a connector and paste one URL: https://app.reviewflowz.com/mcp/tools. ChatGPT takes remote servers only, and this is one, so there is nothing to install on your machine and no local process to keep running.
ChatGPT sends you to Reviewflowz to sign in and shows you what is being granted. OAuth 2.1 with PKCE, so no API key is generated, pasted into a settings box, or left in a password manager for two years. Under five minutes end to end.
Add the connector about four minutes Turn on developer mode
Settings, Apps & Connectors, Advanced settings.
Create a connector and paste one URL
Sign in and approve read-only access
Reviewflowz shows you the grant. You approve it.
Ask it something
It calls the tools by name, not search and fetch
ChatGPT’s earlier custom connectors could reach exactly two tools, called search and fetch. That is enough to look things up in a knowledge base and not enough for a question with a date range and a rating floor in it. Developer mode dropped the restriction.
So it calls ours: list_review_profiles to find what you monitor, get_reviews for the reviews themselves, get_stats for the arithmetic. Three tools, listed by name in the connector panel, so what ChatGPT can reach is something you can read rather than infer.
The arguments are where the afternoon used to go. get_reviews filters on review_profile_ids, start_date and end_date, a rating floor or ceiling or an exact set, keywords across the title and body, topics, tags, language, country and replied.
Every one of those works on get_stats too, which is why a number and the reviews behind it come from the same question. The model reads the reviews for the why. The server counts them for the how many, out of the database, not a token at a time.
Reviewflowz
3 tools · read only
You describe it, ChatGPT builds the query
Nobody is going to learn a filter language for a question they will ask once. You say what you want the way you would say it to a colleague who knows the account, and the client picks the arguments out of the sentence.
You never see the arguments unless you want to, and you always can. ChatGPT shows the call it made underneath the answer, so a sentence it read too narrowly is visible there rather than silent, which is the difference between an answer you can trust and one you have to.
Being wrong is cheap, which is the part that matters. A misread SQL query costs a rerun and a wait for whoever owns the warehouse. A misread sentence costs a sentence, and you are still in the same conversation when you fix it.
And nothing accumulates. There is no schema to learn, no saved view to maintain, no report to retire when the question changes. What you can ask about is not a list somebody curates, it is whatever you can say about reviews.
“only the ones nobody replied to”
replied: false
“since the March release”
start_date: "2026-03-01"
“the angry ones, in German”
rating: 2 · language: "de"
replied: false · start_date: "2026-03-01"
rating: 2 · language: "de"
One call, assembled from a sentence. You never opened a filter menu.
The second question is the one that matters
A dashboard answers the question it was built for. The useful question is almost always the next one: is this new, was it worse in the spring, is it one platform or all of them, did the fix we shipped in April change anything.
In a chat you just ask. The client goes back to the server with different arguments, and the second answer is built the same way the first one was, out of rows, rather than out of what it said to you a minute ago.
So “was it like this before March?” is a second sentence, not a second project. get_stats takes a start_date, an end_date and a period of day, week or month, and hands back the series instead of an impression.
That is the difference between a report and a conversation. A report is finished when it is printed. This one is finished when you stop being curious, which is usually two or three questions further along than you expected.
Was it like this before March?
start_date: "2025-12-01" · period: "month"
December to February averaged 19 a month. March onward averages 67.
The first answer never said when it started. Nobody would have run a second report to find out.
Read-only, in the mode built for writing
Developer mode exists so that a connector can act, and ChatGPT says so on the way in. The warning it shows about unverified connectors is the right warning to give, and anyone adding one should read it. It has nothing to bite on here.
The Reviewflowz server exposes three tools and all three read. There is no fourth. Nothing on it creates, edits or publishes, and that is not a switch left in the off position, it is the entire surface of the server.
The screen you approve says as much: read your review profiles, read your reviews. There is no write permission to grant because there is no tool that would use one, which makes it the rare permission list you can check by reading it rather than by trusting it.
And the connection is yours rather than the account’s. It runs with your permissions, it is revoked from your side, and it never becomes a standing key that outlives whoever set it up. Replies still go out from Reviewflowz, Slack or your helpdesk, where your rules decide who may send one.
Developer mode Reviewflowz connector Custom connectors are not verified by OpenAI. A connector that can take actions can be talked into taking one.
There is no action to take.
Ask the question you kept dropping
One URL, read-only, every review you monitor. Set up in under five minutes.
What do I need on the ChatGPT side?
Developer mode, which requires a paid plan. Plus, Pro, Business, Enterprise and Edu can enable it under Settings, Apps and Connectors, Advanced settings. The free plan cannot add custom connectors. On a Business or Enterprise workspace an admin turns developer mode on before it appears for anyone else.
What do I paste in?
One URL: https://app.reviewflowz.com/mcp/tools. Name the connector Reviewflowz, save, and ChatGPT sends you to a Reviewflowz screen to authorize read-only access. There is no API key anywhere in the flow, and nothing to install locally, because it is a remote server.
Do I have to tell it to use Reviewflowz?
Usually yes, at least at the start of a conversation. ChatGPT will not always reach for a custom connector on its own, so naming it in the question is the reliable way to get an answer built from your reviews rather than from general knowledge.
How much history can it see?
All of it. The connector queries your account rather than reading something you pasted, so it is not limited to what fits in a context window, and a question about last year is the same size of question as one about last week.
Can it reply to reviews for me?
No. The three tools are list_review_profiles, get_reviews and get_stats, and all three read. There is no tool that writes, so there is nothing to switch off. Replies go out from Reviewflowz, Slack or your helpdesk, where your access rules apply.
Can I connect Claude as well?
Yes. It is the same server URL and the same three tools, and each client holds its own authorisation, so connecting one does not affect the other and revoking one leaves the other working.
Do I need a demo to get started?
No. You connect your review profiles and add the connector yourself, in under five minutes. 14 day free trial, no credit card.