How does MCP Analytics choose the right analysis for my data?

MCP Analytics creates a semantic fingerprint of your data by analyzing column types, value distributions, and structural patterns. It then matches this fingerprint against its validated statistical analyses using embedding similarity. Your natural language question further refines the match by mapping intent to analytical concepts. The tool with the highest combined compatibility score runs automatically.

What are semantic embeddings and why do they matter for analytics?

Semantic embeddings convert data descriptions and questions into numerical vectors that capture meaning, not just keywords. This means asking 'Why are customers leaving?' matches to churn prediction tools even though the word 'churn' was never used. It's how the platform understands intent rather than relying on exact keyword matches or manual tool selection.

What happens to my analysis results after they're generated?

Every analysis result is embedded into a searchable knowledge graph. Metrics, trends, segments, and insights from each report are converted to semantic vectors and connected to related findings across all past analyses. This means a question like 'What do we know about churn?' surfaces relevant insights from customer analytics, finance, product, and operations — automatically connected.

Do I need to know which statistical test to use?

No. The platform automatically selects the appropriate statistical method based on your data structure and question. If you have time-series data and ask about forecasting, it picks Prophet or ARIMA. If you have categorical outcomes and ask about drivers, it selects logistic regression or random forest. You describe the business question; the platform handles the statistics.

How is this different from ChatGPT or other AI tools for data analysis?

ChatGPT generates code that you then need to run, debug, and interpret. MCP Analytics runs actual statistical computations on your data using validated R packages, produces interactive visualizations, and generates reports with proper statistical validation (p-values, confidence intervals, effect sizes). Results are reproducible and shareable — not generated text that may hallucinate statistics.

From a plain-English question to an analysis you can defend.

Upload. Ask. Know. Here’s exactly what happens in the minutes between your question and a validated, citable report you own — and why the numbers hold up when someone checks them.

In your AI chat

Connect once in Claude, ChatGPT, Cursor, or any MCP agent — then just talk. Your assistant uploads the data, commissions the analysis, and hands you the report link.

2-minute install →

In the web app

No AI client needed — a guided assistant at account.mcpanalytics.ai walks you from CSV to commissioned analysis in one flow.

Start in the browser →

Step 1

Bring your data

Upload a CSV or connect a live source. Your data is encrypted and private — never shared, never used to train models.

CSV upload
Drag and drop any CSV. Column types and sample rows previewed instantly.
Live connectors
Google Analytics 4, Search Console, Shopify, Stripe — analyze live data without exporting.
Demo datasets
Try everything first on sample data — marketing spend, retail demand, survey responses.

Step 2

Ask your question. Pick a depth.

No method names, no menu of tests — just what you want to know, in your own words. An analyst-style assistant sharpens the question, then you choose how deep to go. The price is always shown before you run.

Plain English in
“Which customers are most likely to churn, and why?” is a complete specification here.
Three depths
A fast verified read, a presented one-page answer, or the full commissioned study.
No surprise billing
Fixed price per depth. A build that fails is never billed — ever.

Create analysisprice shown before you run

Which customers are most likely to churn, and why?

Snapshot — a fast verified read $1 · ~2 min

Brief — the answer on one page SELECTED $5 · ~7 min

Deck — the full commissioned study $25 · 30–45 min

500 welcome credits cover your first Brief.

Step 3

A pipeline of specialists builds your analysis

Nothing here is a chatbot improvising. Your question is scoped to the right statistical method, written as real R, executed on your actual data in an isolated container, and independently checked before you ever see it.

The same pipeline runs at every depth — a Snapshot runs a lean pass in minutes, a Deck runs the full study. Depth changes how far it goes, never the rigor. See the build pipeline in detail →

Build timeline — Brief~7 min total

Scope the method

Data shape read; logistic regression chosen — matched, not guessed

Write it as real R

Deterministic code, fixed seeds — readable, re-runnable

Run on your data

Isolated container; assumptions checked (VIF, fit, sample size)

Independent verification

A separate pass reviews the numbers and the narrative ✓ passed

Delivered

Report + method + citation — billed only now, on success

Step 4

It’s yours. Re-run it forever.

Every analysis you commission becomes a durable module registered to your account — not a chat transcript. Next month’s data? Same analysis, half the price, no rebuild.

Re-run at half price
Fresh data through the exact same validated method — a Brief re-runs for $2.50, a Deck for $12.50.
Share & cite
Token links with no login to view; one-click APA, MLA, Chicago, BibTeX.
The R source, included
Reproduce it independently — same input, same answer, every run.

Your modulesaccount.mcpanalytics.ai

Churn drivers — logistic regression BRIEF

built Jun 03 · last run Jul 01 · subscriptions.csv

Re-run on new data · $2.50 Share Cite

Every module you build lands here — permanent, searchable, yours.

Step 5

Every report makes the next one smarter

Your analyses are remembered by meaning. 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?”

Churn drivers — Brief, June 92% related

Cohort retention — re-run, July 85% related

…the rest stays quietly out of the way.

How the knowledge layer works →

Ask your first question

500 welcome credits on signup — your first Brief is on us. No card required, and a failed build is never billed.

Start free See a live report