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.
Cymple
Data Scientist
Send me your data and question, I’ll send you the analytics.
Custom analysis, built for your CSV.
Upload your CSV, describe your question, and we build a custom statistical analysis module specific to your business: revenue trends, churn signals, ad performance, forecasting. Yours to own. Rerun next month on fresh data without rebuilding.
Real R, not vibes-based AI.
Validated statistical methods (t-tests, ANOVA, regression, survival analysis, PCA) packaged into a custom module for your data. Every result includes diagnostics, confidence intervals, and a citable PDF. Rerun the exact same analysis next semester on next year’s data.
Your AI already knows us.
Connect from Claude, ChatGPT, Cursor, Windsurf, or your own agent via MCP protocol. Full toolset for module discovery, execution, and customization. Bring your own API key or use platform credits. The analysis engine works regardless of which model you use.
What you get back
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.
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.
Ask one question across everything you have ever had analyzed, and the answer cites the reports it came from.
Cymple, data scientist
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.
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.
Try it on your own data
14 days free.
The whole platform, every depth. No card, no call, nothing to install.
Cymple’s card, both sides. Hand it to someone who has a spreadsheet and a question.
See it
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.
What you own
Who it’s for
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 →
How it works
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.
Chosen for the shape of your data and the question you asked, from the standard families. Named in the report, never a black box.
Real R against your file, not a template with your numbers dropped in. The code ships with the answer.
The result is verified independently before delivery. A build that fails its own checks is never billed.
Re-run it next month on fresh data, or modify it. The analysis is yours, at whichever depth you chose.
The standard library
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.
What the AI does
Each card in your report gets its own AI interpretation. Not generic summaries: specific observations about your data, statistical significance calls, and actionable recommendations.
AI reads each chart and table, explains what matters and what to act on.
One-page TLDR for stakeholders who won't read the full report.
Citable R code, assumption checks, what the test actually proves.
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%.
Ask it in plain English
Describe what you want to know. The agent picks the right analysis and delivers a full report.
The memory
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.
Ask a question, pick a tool, or let the agent decide. The result is a full report with charts, tables, and AI insights.
Every result is stored by what it means, not what it was named, so related work finds each other automatically.
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
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.
Chat threads forget. Notebooks scatter. Here, the churn study you ran in May is part of the answer you get in September, automatically.
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
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.
# 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
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.
Every analysis is permanent and shareable by link. Reports kept 365 days, findable months later by what they said, not what you named them.
Methodology and R source on every report, in APA, MLA, Chicago or BibTeX. A paper trail you can hand to anyone who asks.
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.
Connect your sources once instead of exporting a CSV every time. The analysis reads from where the data already lives.
Put a report on a standing schedule and the Monday number is waiting for you, delivered without asking again.
A dataset that does not quite fit the library gets a custom build instead. The unusual question is the normal case here.
The alternative
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.
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
Eight families of validated R analysis. Every one produces a full interactive report with the method named and the code attached.
t-test, ANOVA, ANCOVA, chi-square
linear, ridge, lasso, logistic
random forest, XGBoost, k-means, PCA
ARIMA, seasonality, changepoint
RFM, churn, cohort retention
ROAS, attribution, spend efficiency
Cox PH, uplift, counterfactuals
catalogue mix, basket, cancellation
Start without an account
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.
The guarantee
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.
Start here
Every analysis you run becomes searchable knowledge. The more you use it, the smarter your organization gets.
14 days of full access when you finish onboarding · no card required
Tell us what went wrong, in your own words. We capture the page you're on automatically, so no need to describe where you are.