Statistics & figures · desktop, CLI, REST API, MCP · version 0.2.2

From raw table tofinished figure,in one program

AnaGraph brings data import, descriptive statistics, hypothesis testing, regression, survival analysis, clustering and a large plot library into a single program. No more moving between a script editor, a plotting tool and an office suite to get from raw data to a finished figure.

Request a licence or trial See what it does

Runs locally on your machine. Your data stays there — nothing forces it into a cloud.

Statistical procedures: 30+
Plot types: 29
Tools for automation: ~145
Front-ends, one engine: 4
UI languages: 8

The procedures a lab actually needs

Every analysis runs in the analysis studio: assign columns to the Y, X and group roles by drag and drop, pick the procedure, run it. If you are unsure which test fits, a decision tree gets you there in two or three steps — with the reasoning shown.

Comparison tests

Regression

Multivariate, clustering, categorical

Diagnostics, agreement, survival

29 plot types, one builder

In the plot builder you drag variables into the Y, X, colour/group zones and the facet zones, pick a type and see a live preview. Every figure can then be refined in the inspector down to the axis labels, and exported as PNG or as a Plotly specification.

Grouped scatter plot — petal and sepal measurements of three iris species — **Grouped scatter plot** — petal and sepal measurements of three iris species

Scatter plot with trend line — trunk volume over girth, linear trend line — **Scatter plot with trend line** — trunk volume over girth, linear trend line

Scatter plot matrix — every pair of measurements at a glance — **Scatter plot matrix** — every pair of measurements at a glance

Line chart — dose-response per preparation — **Line chart** — dose-response per preparation

Area chart — the same trend as a filled area — **Area chart** — the same trend as a filled area

Box plot — quartiles and outliers per species — **Box plot** — quartiles and outliers per species

Violin plot — density shape per species — **Violin plot** — density shape per species

Beeswarm plot — every single observation as a point — **Beeswarm plot** — every single observation as a point

Ridgeline plot — distributions stacked above each other — **Ridgeline plot** — distributions stacked above each other

Histogram — sepal length in bins — **Histogram** — sepal length in bins

Density plot — smoothed kernel density per species — **Density plot** — smoothed kernel density per species

2D density plot — joint density of two measurements — **2D density plot** — joint density of two measurements

Correlation heatmap — Pearson correlations across the dataset — **Correlation heatmap** — Pearson correlations across the dataset

Bar and scatter combination — mean per dose plus the raw data — **Bar and scatter combination** — mean per dose plus the raw data

Stacked bar chart — composition per dose — **Stacked bar chart** — composition per dose

Lollipop chart — dry mass per treatment — **Lollipop chart** — dry mass per treatment

Circular bar chart — the same values arranged radially — **Circular bar chart** — the same values arranged radially

Pie chart — displacement by cylinder count — **Pie chart** — displacement by cylinder count

Treemap — area proportional to share — **Treemap** — area proportional to share

Waffle chart — shares as a grid of cells — **Waffle chart** — shares as a grid of cells

Parallel coordinates — one line per flower across four axes — **Parallel coordinates** — one line per flower across four axes

Radar chart — measurement profile per species

Dendrogram — hierarchical cluster structure (Ward) — **Dendrogram** — hierarchical cluster structure (Ward)

Every figure was rendered with AnaGraph from datasets the program ships with or validates against — the application's own example dataset and the classic R datasets from the validation suite. Click to enlarge.

Points & lines

Distribution

Proportions, areas & specialty

Four ways into the same engine

Statistics, plotting, import and persistence live in separate, front-end agnostic libraries. The desktop application, the command line, the REST API and the MCP server all call the same tool registry — you get the same number whichever way you ask. That is the difference between "we also have an API" and "the API is the same program".

Desktop application

The daily workspace: data table, analysis studio, plot builder, output document. Packaged for Windows and macOS with automatic updates; the interface speaks eight languages.

Command line

For batch runs, cron jobs and CI pipelines: run a finished analysis reproducibly against new data, with no screen attached.

statorium analyze \
  --recipe descriptive-overview \
  --data study.csv \
  --output report/

REST API

An HTTP interface for dashboards, reporting tools and your own scripts. Described by an OpenAPI document, with Python and R snippets for getting started in a notebook.

MCP server

Exposes the engine over the Model Context Protocol to LLM clients such as Claude Desktop or Cursor. The difference to an assistant without it: it computes the t-test instead of suggesting code for one.

Automation

AnaGraph speaks MCP

The bundled MCP server exposes the complete tool inventory over the Model Context Protocol. Claude Desktop, Cursor or your own agent connect as clients — and get a statistics engine rather than statistics advice. The difference shows on the first t-test: an assistant without it suggests code someone has to run. With AnaGraph it computes.

You ask in the chat

“Take the iris dataset and show me sepal length as a box plot, split by species.”

The client calls a tool

Not free-form code — a schema-bound tool from the registry:

createPlot {
  "datasetName": "Iris",
  "plotType":    "BoxPlot",
  "mappings": {
    "y":     "Sepal.Length",
    "group": "Species"
  }
}

You get the figure

Box plot of sepal length per iris species, produced by the tool call shown — This exact call produced this exact image.

Scope

Around 130 tools over MCP

Not just figures: import, reshaping, filtering, every analysis, output and export are available as tools — the same inventory the chat panel and the REST API expose.

Transport

Local or over HTTP

The default is stdio — client and server on the same machine, nothing leaves the device. Where the client runs elsewhere, the server can listen as an HTTP endpoint (streamable HTTP) instead.

Without a language model

Command line and REST API

The same tools without an LLM: the CLI for batch runs and CI pipelines, the REST API for dashboards and your own scripts. A saved recipe later runs unchanged against new data.

From import to report

Between raw file and figure sit the steps that actually cost time in daily work. They are built in, not bolted on.

Import

CSV, Excel — and Prism projects

Drag a CSV onto the window, open Excel workbooks directly. GraphPad Prism project files (.pzfx, .prism, .pzf) are read with their data tables and their stored analyses; Prism layouts are reshaped to long format on the way in.

Output

One document, five formats

Tables, statistics and text from an analysis live in one output document and export as Word, Excel, HTML, Markdown or plain text. Figures export in bulk as PNG or as Plotly specifications.

Repeatability

Command log and recipes

Analyses, data transforms, plots and imports are recorded and can be replayed — with a JSONL audit log alongside. Saved as a recipe, the same sequence later runs from the command line against new data.

Cross-check

Second opinion

An audit pipeline re-checks a finished analysis against the data: outliers, model assumptions, a candidate transformation, a distribution-free alternative, influential points. It does not replace your judgement — it shows which assumption is shaky.

Checkable, not merely claimed

Statistics software is worth exactly as much as the trust in its numbers. So rather than asserting that it computes correctly, here is what it is checked against.

Against reference implementations

The project ships a validation suite that computes the procedures against established Python references — scipy, statsmodels, numpy, lifelines — and records the expected values per analysis feature. When an algorithm changes, the difference shows up before it ships.

Against the NIST datasets

For linear and nonlinear regression and for analysis of variance, comparisons against the NIST certified datasets are on file — the classic benchmark problems with known reference values used to measure numerical accuracy.

Against your existing analysis

On Prism import, the headline statistics — R², AUC, standard error, confidence interval, p, df, N — are placed side by side with the values stored in the Prism file. You see on your own data whether the numbers agree, before you switch.

With the source named

The bundled help cites the primary literature with DOIs for each procedure — for the mixed model, Henderson (1975) and Laird & Ware (1982), for instance. If you have to cite a method in your methods section, the reference is in the program.

Three properties that make the difference

There is no shortage of statistics programs. These three points are why AnaGraph is used in regulated and privacy-sensitive environments — and all three can be checked before you buy.

Local

The computation runs on your machine

AnaGraph is a desktop application, not rented access to someone else's data centre. Import, analysis and figure all happen on the machine the program runs on. No step of the work requires an internet connection.

Data ownership

Your measurements never leave the building

If you work with patient data, batch records or unpublished measurement series, you have to be able to say where that data sits. With an application that computes locally the answer is short: where you put it. That saves a data processing agreement and a third-country transfer assessment.

Automatable

The same computation with no screen

Whatever you set up once in the program then runs unchanged in a pipeline, a cron job or from an LLM client — through the same tool registry and with the same result. A statistics program you can only drive with a mouse cannot do that.

Who is behind it

The vendor and point of contact for AnaGraph is STATCON GmbH, based in Kassel, Germany. For more than 20 years STATCON has supported companies in pharma, chemistry and materials science on statistical questions — consulting, training and software.

People, not a ticket system

You write to one address and get an answer from someone who knows the method — in German or English. Technical questions about a procedure are explicitly welcome, including before you buy.

Training and consulting alongside

Software does not replace methodological competence. Where that is missing, training and statistical consulting are available from the same house — on request also on site.

Statistics is the business, not a sideline

For more than 20 years STATCON has sold, taught and consulted on statistical methods and nothing else. Calling here does not put you through to a general software sales desk but to people who work with the same procedures you do.

Licensing and procurement

AnaGraph is commercial software and requires a licence key. A trial with the full feature set can be started from inside the program; activation then happens in the licence dialog.

Prices are not published here yet

There is currently no published price list for AnaGraph. A number here would be a commitment nobody has made yet, so there is none. What does apply today:

Quotation on request, including against a purchase order and by invoice.
Single and multi-seat licences are possible; we clarify the scope in conversation.
Academic and teaching use is normally priced separately.

Tell us how many seats and what the intended use is — you get a firm quotation rather than a ballpark figure.

The contractual framework exists in draft form: the terms and conditions cover ordering, term and liability, and the end user licence agreement covers the right of use at the workstation (named user). Until legal approval, the individual quotation governs.

Get in touch

There is deliberately no contact form. An email can be forwarded, filed with a procurement record and read by several people — a form cannot.

Trial access, a quotation, a demonstration or a technical question about a procedure — write to

vertrieb@statcon.de

or by phone +49 5542 93-300