★ Open Beta · 100% Free · No Card Required

GA4 MCP for Claude, ChatGPT, Cursor, Perplexity.

No sampling. 90 days preloaded. Connect in two minutes.

One OAuth click. One connector URL. Paste it into Claude, ChatGPT, Cursor, Perplexity, Telegram — start asking your GA4 questions in plain English.

No row caps. No sampling. 90 days of your data cached into Postgres on day one — no waiting for the API, no rebuilding queries between sessions.

Works in

ClaudeChatGPTCursorPerplexityTelegramWeb

Get the connector URL →

Open beta — completely free. No card required.

LIVEDemo ShopMessages: 0 / 5Off-topic: 0 / 3

Compare the last 7 days to the previous 7. What changed?

Pulling overview metrics

Mapping funnel

Last 7 days vs previous 7 — Demo Shop

Metric	This week	Last week	Δ
Sessions	6,914	7,512	−8.0%
Conversions	51	78	−34.6%
Revenue	$6,724	$9,810	−31.5%
CR	0.74%	1.04%	−28.8%

Where it broke: the drop is concentrated in add_to_cart → begin_checkout (70% → 56%). Traffic mix is steady — this is a checkout-step regression, not an acquisition problem.

Ask me to dig deeper — e.g. "why is checkout dropping?" or "is this device-specific?"

How it works

Sign up. Paste one URL. Start asking.

Three steps. The longest one is waiting for the initial sync.

Sign up & connect GA4

Email and one Google OAuth click. Read-only access, revoke anytime. Pick the GA4 property — we'll do the rest.

~2 minutes

Paste the connector URL

One URL works everywhere — Claude (web, Desktop, Cowork), ChatGPT, Cursor, Perplexity, Telegram. No per-platform config, no service accounts, no IAM.

~30 seconds

~1 hour

Initial sync & 90-day preload

Ask your data

We preload 90 days of GA4 into Postgres so your first query returns instantly. Plain English in, structured answer out — with the SQL we ran, if you want to see it.

~1 hour for initial sync · queries instant after that

Why the answers come back fast

Your GA4, in Postgres. Queries run as code.

Most MCPs hit the GA4 API on every question — sampled, rate-limited, slow on anything beyond 14 days. ConvRadar ingests your data nightly into our own Postgres, so queries run as SQL against full history. The LLM writes the sentence after the data comes back, not before.

Your GA4, ingested nightly

Pulled into fact tables on our Postgres. No live API calls. No sampling. No row caps. 90 days available from minute one, full history available within a week.

Queries run as SQL

When you ask “show me 90-day conversion rate by source,” we run actual SQL against your data, not a hallucinated response. The MCP exposes 20+ structured tools — funnel, anomaly detection, segment comparison, cohort retention — all running as code on our server.

The LLM writes the sentence

It receives the finished result and turns it into the answer. The reasoning happens before the writing.

Read-only OAuth. Every number traces back to the GA4 slice it came from.

What you can ask

What you can actually ask. In plain English.

No SQL. No GA4 Explorations. No “filter by event_name”. You ask, we pull the right slice and answer.

“Show me daily sessions for the last 90 days, by source.”

Returns clean time series, all sources broken out, no sampling warning. Same query against native GA4 API gets sampled past 14 days.

“Which landing pages converted best on mobile last month?”

Ranked list with sessions, CR, revenue per page. Filterable, exportable as table.

“Compare this week to last week — what changed?”

Full funnel delta, channel-level deltas, the segment that moved most. Three queries collapsed into one answer.

“Run anomaly detection on conversion rate for the last 60 days.”

Dates where CR deviated >2σ from baseline, with the segment responsible. Detection runs as Python, not LLM guess.

“What’s my retention curve for users acquired in March?”

Cohort table with day-1, day-7, day-30 retention. Computed on our Postgres, not GA4 Explorations.

“I deployed a change on Mar 14 — did it move the funnel?”

Pre/post comparison with confidence interval, by step. “Significant lift”, “no effect”, or “you broke something” — it’ll say which.

“Run a full audit.”

300+ structured checks across traffic, funnel, products, devices, anomalies. Verdict + evidence per finding.

What the right answer actually looks like

Five real findings. Names removed, numbers real.

Every row is the output of a single question — asked in Claude, ChatGPT, Telegram, or right here in the page chat. Verdict, evidence, recommended action — paste it into a Slack thread and stand behind it.

Industry	Problem we found	Fix we shipped	Recovered
DTC apparel	Mobile checkout CVR 38% below desktopLeak localised to the address step on iOS	Replaced address autocomplete widget, removed a blocking JS error in the ZIP validator.	+$42,800/momobile CVR +51%
SaaS	Sign-up to activation drop at the “invite teammate” stepAnomaly started the day after a deploy; only affected free plan	Made the invite step skippable; activation recovered within 48 hours.	+27% activation~$9,100 MRR added
Beauty e-com	Paid search CVR collapsed on one campaignIntent mismatch — page promised “gift sets”, campaign bid on “refill”	Split the ad group, pointed refill queries at the correct PDP, paused 4 mismatched keywords.	+$18,400/mopaid CAC −23%
Marketplace	Top-10 SKU with above-average traffic, below-average CVRProduct images missing; variant picker broken on mobile	Re-uploaded photography, rebuilt the variant picker as a native select on mobile.	+$11,600/moSKU view→cart +64%
Subscription	Organic traffic grew 22%, revenue flatNew traffic landed on a blog post that didn’t link to the offer	Added an inline CTA block and an exit-intent offer; kept the editorial tone.	+$6,900/moblog→trial +3.1pp

Numbers rounded. All figures verified before/after with the ConvRadar confidence interval attached.

FAQ

Questions, answered before you ask.

Which clients does the connector URL work with?

Any MCP-compatible client. Confirmed working: Claude.ai (web, Desktop, Cowork), ChatGPT (any paid plan), Cursor, Perplexity, Telegram. One URL across all of them — no per-platform config.

How is this different from connecting GA4 to Claude directly?

Generic GA4 connectors hit the live API on every question — which means sampling on anything past 14 days, 10k row caps, and seconds of latency per query. ConvRadar pre-ingests your GA4 into our own Postgres, so 90 days are available instantly with no sampling, ever. Plus 20+ structured analytics tools (funnel, cohort, anomaly detection) that the raw API doesn’t expose.

Do I need a Claude or ChatGPT account?

Only if you want to use it from there. ConvRadar runs as an MCP connector inside Claude.ai / Desktop / Cowork and ChatGPT (any plan). If you’d rather skip both, it also works in Cursor, Perplexity, Telegram, or right here on the webpage.

Does it support OAuth 2.1 with Dynamic Client Registration?

Yes. The MCP server at mcp.convradar.com implements OAuth 2.1 + DCR. Compatible with Claude, ChatGPT, Cursor, and any other client that follows the MCP auth spec.

Can I see the SQL the tool ran?

Yes. Every tool call writes to an audit log — the SQL, the parameters, the rows returned. Pull it up anytime to verify a number or hand it to a teammate.

What GA4 entities are supported?

Sessions, events, conversions, e-commerce (items, transactions, revenue), traffic sources, landing pages, devices, geo, custom dimensions, audiences, attribution. If GA4 tracks it, the MCP exposes it.

Is there a rate limit?

No. Because queries run against our local Postgres, not the GA4 API, you’re not hitting Google’s quota. Ask as many questions as you want.

What about my data — is it safe?

GA4 access is read-only via Google OAuth. We never write to your property, never see your raw user-level events outside the metrics GA4 already aggregates, and you can revoke access in one click. Numbers used to answer a question stay in the audit log so you can see exactly what was pulled.

Does this work for SaaS, or only e-commerce?

Both — plus lead-gen, marketplaces, and subscription apps. The funnel logic is adaptive: e-com is view → cart → checkout → purchase; SaaS is signup → activation → upgrade; lead-gen is visit → form view → submit. ConvRadar reads what your GA4 actually tracks.

How long until I get a useful answer?

Ten minutes to your first answer. Five minutes to sign up and OAuth GA4, one minute to paste the connector URL into your client of choice, then ask. The 90-day initial sync runs in the background — most queries work within the first few minutes, full history within an hour.

How much does it cost?

Free during open beta. Sign up with your email — no card, no checkout, no time limit while the beta is open. We’re collecting users and feedback before launch.

Can I cancel anytime?

Yes. Monthly subscription when we leave beta, one click to cancel, no questions, no retention call.