Works inside Claude, ChatGPT & any MCP agent

Statistical analysis you can actually defend.

Ask a question about your data in plain English, right in your AI chat. Your chatbot gives you an answer. MCP Analytics runs real statistical computation and gives you the number you can publish: the method named, citable, and yours, with next month’s number delivered without asking again.

SimpleRepeatableDefendableSearchable
14 days of full access when you sign up, subject to fair use. No card required.
Every depth unlocked for 14 days, fair use Failed builds never billed Your analyses stay yours to run

Cymple

Data Scientist

Send me your data and question, I’ll send you the analytics.

ds@mcpanalytics.aimcpanalytics.ai
See how we compare to plain AI →

What you get back

An analysis, not an answer.

Repeatable

The same code on the same numbers returns the same numbers. Send next month’s file and the shape holds, which is what makes this quarter comparable to last.

Defendable

You get the method it used and the code that ran it, not just the number. When someone asks how you know, the answer is in your hands.

Searchable

Ask one question across everything you have ever had analyzed, and the answer cites the reports it came from.

Cymple, data scientist

Send a question and a spreadsheet.

Cymple picks a standard method, writes the analysis as real code, runs it on your numbers, and sends back the answer, the method, and the working.

  • Question“Which of my channels actually paid for itself last year?”
  • Datachannels-2026-h1.xlsx

That’s it. You get the analytics back.

Fourteen days of full access, no card. Subject to fair use.

Cymple

Simple · Repeatable · Defendable · Searchable

Cymple

Data Scientist

Send me your data and question, I’ll send you the analytics.

ds@mcpanalytics.aimcpanalytics.ai

Try it on your own data

14 days free.

The whole platform, every depth. No card, no call, nothing to install.

mcpanalytics.ai

Cymple’s card, both sides. Hand it to someone who has a spreadsheet and a question.

See it

Thirty seconds, or the whole argument.

There is also a method series. A/B tests, survival curves, multiple comparisons, sampling. What each method answers, when it misleads, and how our agents choose between them.

Or just write to Cymple

Attach the spreadsheet, put the question in the body. No account, no form, nothing to install.

ds@mcpanalytics.ai

What you own

A piece of work, not just a report.

Calculated, not hallucinated: R computes the numbers on your data before a word of insight is written.
Hourly Conversion: Ads vs Control Brief · computed rates · interactive
A delivered Brief: hourly conversion rate, ad treatment vs PSA control on a public marketing A/B dataset: two-line comparison chart, computed rates, and AI-written insight
In every deliverable

Everything a defensible answer needs.

How the work gets built →

Who it’s for

One of you, or all of you.

Two plans, and you already know which one you are. A person takes Academic. A company takes Business. Everything else about them is the same.

One person

Academic $19 /mo

Sized for you and your own data. Custom analyses built for you, and yours for good. Add more any time, and nothing you buy expires.

A company

Business $449 /mo

Sized for a team and everything they ask. A specialist builds your first reports with you. Deep commissioned studies, seats, a priority queue and standing schedules.

By contract

Enterprise talk to us

The platform under your brand, on your servers, with SSO and audit logs. Pipelines customised to your business.

The plan never decides whether the work was done properly. Every depth of analysis, every connector, scheduling, your library and its search, and the method and the R source on every report, are in all of them. What changes is how much work it is sized for.

Neither shape fits? Set your own monthly limit from $85 and change it whenever, or pay per analysis with no subscription at all. Compare everything →

Start with 14 days free, no card →

How it works

Built to order, for your question.

You send a question and a file. The pipeline picks a standard method, writes the analysis as real R, runs it on your numbers and checks its own result before you ever see it.

  1. 01

    It picks the method

    Chosen for the shape of your data and the question you asked, from the standard families. Named in the report, never a black box.

  2. 02

    It writes the code

    Real R against your file, not a template with your numbers dropped in. The code ships with the answer.

  3. 03

    It checks itself

    The result is verified independently before delivery. A build that fails its own checks is never billed.

  4. 04

    You keep it

    Re-run it next month on fresh data, or modify it. The analysis is yours, at whichever depth you chose.

Understand the pipeline →

The standard library

Or skip the build: run a prebuilt analysis.

A library of platform-built, independently verified analyses covering the workhorses of applied statistics. No build wait: bring your data, map your columns, get the full report in minutes.

Every library analysis was built by the same pipeline that handles custom work, then verified and published. Each one is a complete, multi-section statistical report: the method, the checks, the charts, the plain-language findings. It runs on your data the way it ran on ours, and it's filed by the question it answers, so you find it by what you want to know, not by remembering what the method is called. Whatever you run joins your own library, yours to run again whenever the data changes.

A/B testing Cohort retention Forecasting Key drivers Clustering Anomaly detection Linear regression Correlation Group comparison Market basket Control charts Factor analysis Event impact Data profiling …and more
Browse the standard library

What the AI does

AI Insights on Every Card

Each card in your report gets its own AI interpretation. Not generic summaries: specific observations about your data, statistical significance calls, and actionable recommendations.

Card-level analysis

AI reads each chart and table, explains what matters and what to act on.

Executive summary

One-page TLDR for stakeholders who won't read the full report.

Methodology notes

Citable R code, assumption checks, what the test actually proves.

Key Findings: Marketing Spend Analysis

TV spend shows the strongest ROI at $4.20 per dollar, significantly outperforming Radio ($2.15) and Newspaper ($0.87). The model explains 89.7% of variance (R² = 0.897), suggesting marketing budget reallocation from Newspaper to TV could increase revenue by approximately 12-18%.

OLS Linear Regression  |  p < 0.001  |  n=200 High confidence: R² > 0.85
Learn more about AI insights →

Ask it in plain English

Ask Questions. Get Answers.

Describe what you want to know. The agent picks the right analysis and delivers a full report.

account.mcpanalytics.ai
Why are customers leaving after the first month?
C
Cymple matched your data to Churn Prediction
Report ready · 8 cards · 42 seconds
Monthly churn is 4.2%. Users who skip onboarding churn at 3.1x the rate of those who complete it. The first 7 days are critical: 68% of churners never return after day 3.
View full report Export PDF Run cohort retention next
See how the agent works →

The memory

Every Analysis Builds Your Knowledge

Every result gets embedded into a high-dimensional vector space. Related insights cluster together automatically.
Search by meaning, not keywords. The more you analyze, the more connections you find.

Learn More About the Knowledge Layer →

1

Run an Analysis

Ask a question, pick a tool, or let the agent decide. The result is a full report with charts, tables, and AI insights.

2

Remembered by Meaning

Every result is stored by what it means, not what it was named, so related work finds each other automatically.

3

Search by Meaning

Ask “What do we know about churn?” and the system retrieves the most semantically relevant results, across all datasets, tools, and time.

Search by meaning

Ask once. Everything you’ve ever analyzed answers.

Every report you run becomes memory. Ask a new question, and the most related past work surfaces on its own, whichever tool built it, whenever it ran.

You ask “What do we know about churn?”
Your most related past analyses, found by meaning, not by filename:
Churn Prediction DECK Ran in May · subscriptions dataset 92% related
RFM Segmentation BRIEF Ran in March · orders dataset 87% related
Onboarding Funnel BRIEF Ran in June · product events 81% related
…everything less related stays quietly out of the way.

Why this matters

Chat threads forget. Notebooks scatter. Here, the churn study you ran in May is part of the answer you get in September, automatically.

How it works

Results are matched by meaning, so “churn” finds retention cohorts and cancellation drivers even when nothing shares a keyword. The full mechanics →

The more you analyze, the smarter every next question gets.

For developers

Plug Into Your Agent

MCP Analytics is an MCP server. Install it in any compatible AI client and your agent gets access to the full module library: run statistical analyses, ML models, and business analytics directly from conversation.

One developer installs the MCP server. The whole team views reports in the web app. Two interfaces, one platform.

Claude Desktop Cursor Windsurf Claude Code Any MCP Client
See the full integration guide → | API docs
Install via npx
# Add to your MCP client config { "mcpServers": { "mcpanalytics": { "command": "npx", "args": [ "-y", "@mcp-analytics/mcp-analytics", "--api-key", "YOUR_API_KEY" ] } } }

Also supports direct HTTP and OAuth2

What else you get

It does not stop at the answer.

Not a chat thread that disappears, and not a notebook nobody can find. Every analysis is permanent, shareable and citable, and the platform keeps working after the first one.

It stays

Every analysis is permanent and shareable by link. Reports kept 365 days, findable months later by what they said, not what you named them.

It cites

Methodology and R source on every report, in APA, MLA, Chicago or BibTeX. A paper trail you can hand to anyone who asks.

It re-runs

Anything built for you is yours for good. Run it again on next month’s data at half the build price, same code, comparable numbers.

It connects

Connect your sources once instead of exporting a CSV every time. The analysis reads from where the data already lives.

It schedules

Put a report on a standing schedule and the Monday number is waiting for you, delivered without asking again.

It adapts

A dataset that does not quite fit the library gets a custom build instead. The unusual question is the normal case here.

The alternative

The analyst takes days. The chatbot guesses.
We compute, in minutes.

Fluent and correct are not the same thing, and rigorous doesn’t have to mean slow. Real statistical analysis used to be a choice between waiting and guessing. Now it isn’t.

Hiring it out

Days – weeks
  • ~$600 of analyst time per study
  • Back-and-forth to scope the question
  • Static PDF: new data means a new engagement

A chatbot alone

Instant, but unverifiable
  • Numbers generated by a language model
  • ~33% accuracy on inferential statistics with everyday prompts*
  • A transcript: ask again tomorrow, get a different answer

MCP Analytics

Minutes, computed & checked
  • Real R computation on your actual data
  • The right test, justified: assumptions checked, shown
  • A durable analysis you own: refresh it on next month’s data

Microsoft tells Copilot users not to rely on it for work “requiring high accuracy or reproducibility.” We exist for exactly that work.

*accuracy figure: peer-reviewed evaluation of ChatGPT’s statistical analysis, JMIR 2025

The methods

Standard methods, run as real code.

Eight families of validated R analysis. Every one produces a full interactive report with the method named and the code attached.

Hypothesis testing

t-test, ANOVA, ANCOVA, chi-square

Regression

linear, ridge, lasso, logistic

Machine learning

random forest, XGBoost, k-means, PCA

Time series

ARIMA, seasonality, changepoint

Customer analytics

RFM, churn, cohort retention

Marketing and ads

ROAS, attribution, spend efficiency

Causal and survival

Cox PH, uplift, counterfactuals

E-commerce

catalogue mix, basket, cancellation

Browse the library →

Start without an account

Three ways in, none of them a sign-up.

Run a free tool on your own CSV, browse the analyses we have already built, or read how the methods work. No card, no account, nothing to install.

Data Profile: Understand Any Dataset
The “what’s in my data” report: distributions, missingness, and structure at a glance.
Group Comparison: t-test & ANOVA
Does a numeric outcome really differ between groups? The right test, chosen and checked.
A/B Test Analysis
Upload row-level experiment data and get lift, significance, and a decision you can defend.
Time Series Forecast
Forecast any metric from its own history, with honest uncertainty bands.
Correlation Analysis
Pairwise correlations across your numeric columns: what moves together, and how strongly.
Browse all free tools →

The guarantee

If a Report Isn’t Right, We Make It Right.

Every number in every report is computed and independently verified before it reaches you, never improvised by an AI. If a report still isn’t right, flag it: we diagnose, fix and resend at no cost. Failed builds are never billed, and refund requests are reviewed case by case.

🚩
1. Flag the issue
Thumbs down on any slide. Tell us what's wrong: chart broken, wrong data, missing section.
🛠
2. We fix & resend
Our system diagnoses the issue, fixes the module, and sends you an updated report at no cost.
💰
3. Or request a refund
Still not right after the update? Request a refund from the report page, reviewed case by case.
Learn more about credits & refunds →

Keep in touch

Stay in the Loop

Get product updates, new analysis modules, and data science tips delivered to your inbox.

Start here

Start Building Your Analytical Memory

Every analysis you run becomes searchable knowledge. The more you use it, the smarter your organization gets.

Get Your First Report Free See How It Works

14 days of full access when you finish onboarding · no card required