MatPlotLibNet — matplotlib for .NET

MatPlotLibNet is a charting library for C#, .NET 10 and .NET 8, inspired by matplotlib. It draws 83 chart types and saves them as SVG, PNG, PDF or an animated GIF, straight from your server code. It does not need a JavaScript framework, a WebView or a hosted service.

It has a fluent API and it works with dependency injection. It ships with 148 colormaps, 30 themes, 13 map projections with embedded Natural Earth data, LaTeX-style MathText, text in any script, O(1) streaming with 53 technical indicators, a control room (Plt.OpsDashboard() — KPI tiles, state timelines, one shared trend window), and an MCP server so an AI agent can draw with it. Charts render natively in Blazor, WPF, MAUI, Avalonia, Uno and ASP.NET Core, inside a Verso notebook cell, and TypeScript clients are there for Angular, React and Vue.

Where this is going

1.18 draws charts inside a Verso notebook. Verso is the open-source notebook built to replace Polyglot Notebooks, and MatPlotLibNet.Verso is an extension you install from its Extensions panel. After that a cell that ends on a figure — or on the builder that makes one, without .Build() — shows the chart instead of a line of type names. The chart fills the width of the cell, and one taller than the cell can show gets a scroll box rather than being squeezed into a size nobody chose. The package carries the extension and nothing else on purpose: a notebook loads an extension in its own context, so a second copy of the charting library shipped alongside would be a different Figure from the one the cell made. MatPlotLibNet.Notebooks stands still at the version it reached; the runtime under Polyglot Notebooks has ended, and notebook work belongs in Verso now.

1.17.1 is a patch on top of it. A clicked data point now reaches your own code — the desktop controls and the SignalR route both hand it over with its series, its value and where it was. The control room gained a service-topology panel, GroupedBar draws several bars per category in one call, DensityScatter colours a scatter by how crowded it is, and three accessible colour sequences arrived (petroff6, petroff8, petroff10). All fourteen packages carry an icon and release notes now, and the documentation site has a page that translates matplotlib into C#, call by call.

1.17 draws text in any script. Arabic letters join, Hebrew reads right to left, and a number or a Latin word inside Arabic or Hebrew text is placed in the correct position, using the font that ships in the package. The Unicode bidi algorithm is implemented in full and checked against Unicode's own 861,948 test cases; HarfBuzz shapes each run. Other scripts need a font file: SkiaFonts.Register(path) registers one, and the MCP server reads font files from the MATPLOTLIBNET_FONTS variable. Latin text is now kerned too, so text widths changed by up to 2 px per letter pair.

1.16 adds a data table beside every chart, for a reader who cannot see the picture. figure.ToDataTables() returns the figure's data as tables with typed columns, and you can write each table as HTML, Markdown or CSV. Every host serves the table: an ASP.NET Core endpoint next to the SVG endpoint, a ShowDataTable option on the Blazor component, a chartDataTable field in the GraphQL schema, and a fifth MCP tool. The same release fixed the streaming series. A point and a candle are now stored as one value, so a snapshot can no longer show a point or a bar that never existed. The buffer behind them is now the generic RingBuffer<T>.

1.15 added support for AI agents. MatPlotLibNet.Mcp is a Model Context Protocol server. A model can render any of these chart types from a JSON spec and look at the result. When the spec is wrong, the server reports which field is wrong instead of returning a blank picture.

1.14 added the control room: Plt.OpsDashboard() and the tile anatomy arrived with it. The releases since then contain the fixes and extensions that real control-room use required: a secondary Y axis with the same capabilities as the primary axis, streaming that draws in SVG, 3-D axis titles that do not overlap their own labels, tile captions that wrap and stay inside their tile, and an ops window whose ticks are formatted as time.

After that, releases follow what the community needs: bug fixes, documentation, and whatever comes up in real use. Open a Discussion or an Issue; that is what decides the next release. Every release, with its migration notes, is listed in the CHANGELOG.

Quality bar: every class must pass a strict coverage gate of ≥90 % line and branch coverage (706 classes, at 99.6 % line / 97.5 % branch) across 11,757 tests. The test suite verifies rendering against matplotlib pixel-fidelity fixtures.

Documentation

• Wiki — getting started, cheatsheet, all 83 chart types, the control room, styling, streaming, SignalR, packages
• From matplotlib to C# — the pyplot call you know, and the line that draws it here
• Cookbook — copy-paste examples with rendered output
• Playground — try charts live in the browser
• API Reference — generated from the XML docs
• Benchmarks · Coverage policy · CHANGELOG

Packages

Quick start

using MatPlotLibNet;
using MatPlotLibNet.Styling;

double[] x = [1, 2, 3, 4, 5];
double[] y = [2, 4, 3, 5, 1];

Plt.Create()
    .WithTitle("My First Chart")
    .WithTheme(Theme.Dark)
    .AddSubPlot(1, 1, 1, ax => ax
        .Plot(x, y, s => { s.Color = Colors.Blue; s.Label = "Data"; })
        .WithLegend())
    .Save("chart.svg");

Features

83 series types — line, scatter (plain or coloured by point density), bar (plain, stacked, or several groups per category in one call), histogram, pie, box, violin, heatmap, contour, candlestick, OHLC, treemap, sunburst, Sankey, polar, polar heatmap, 3D surface, Bar3D, PlanarBar3D, Line3D, Trisurf3D, Contour3D, Quiver3D, Voxels, Text3D, radar, waterfall, funnel, gauge, stat tile (single-value KPI), state timeline (discrete state segments over time), pair grid, relative rotation graph, streaming line/scatter/signal/candlestick, and more.

MCP server — charts for an AI agent — MatPlotLibNet.Mcp is a Model Context Protocol server. It ships as a .NET tool that a host starts over stdio. It has five tools: render_chart (returns a PNG plus a text summary of what was drawn), save_chart (writes PNG/SVG/PDF to a file and returns the path it wrote), chart_data_table (returns the numbers the picture is drawn from, as markdown, which an image alone cannot give), list_chart_types and describe_chart_schema. The spec is the library's own figure JSON, so an agent can ask for anything the library draws. When a spec is wrong, the server rejects it before rendering and names the wrong field, instead of returning a blank picture. See the cookbook.

{ "servers": { "MatPlotLibNet.Mcp": { "type": "stdio", "command": "dnx", "args": ["MatPlotLibNet.Mcp", "--yes"] } } }

Control room — Plt.OpsDashboard() builds one operator screen: KPI tiles across the top, state timelines under them, a topology panel showing which service calls which (AddTopology), and a shared trend panel. All of them use one time window that the caller supplies; the library never reads the wall clock. A tile carries a Target, a multi-line Caption that wraps, an inline sparkline, and an OpsCondition saying what it is in. That condition has two axes on purpose: OpsSeverity (normal, warning, critical) is the only one that compares, and OpsVisibility (observed, unknown, shelved) says whether the reading can be believed and whether someone silenced it — a silent source is not a degraded one, and a muted alarm is not a solved problem. OpsCondition.RollUp reads a set of children the way an operator does: the worst child that can be seen decides the colour, the unseen and the muted are counted. WithEventMarker puts a deploy or an incident on every panel that has a clock. Theme.Alarm defines the reserved alarm colours, and four operator backgrounds ship with it. BulletGraphSeries replaces the radial gauge.

Native UI controls — MplChartControl for Avalonia 12 and MplChartElement for Uno Platform render charts natively via SkiaSharp. They need no browser, no WebView and no SignalR. There are 9 interaction modifiers: pan (drag), zoom (scroll), 3D rotation (right-drag), rectangle zoom (Ctrl+drag), brush select (Shift+drag), span select (Alt+drag), legend toggle (click), crosshair (passive), hover tooltip. The controls also have a toolbar state model, view history (back/forward), a data cursor (click-to-pin), tick mirroring, and tight margins. Clicking a data point raises DataPointClicked with the point the reader clicked — series, value, pixel position and axes — so an application can open a detail panel or select a row beside the chart. A server-driven chart reports the same click to its OnDataCursor handler.

MathText — LaTeX-like inline math in any label or title: $\alpha^{2}$, $\frac{a}{b}$, $\sqrt{x}$, $\hat{x}$, $\mathbf{F}$, $\mathbb{R}$. It supports 96 symbol mappings (Greek, math operators, arrows, relations, set/logic, blackboard bold), fractions, square roots, accents, font variants, spacing, and scaling delimiters.

International text — labels, titles and legend entries in any script. Arabic and Hebrew work with the bundled font: the Unicode bidi algorithm (UAX #9, all 861,948 conformance cases) finds the reading direction, HarfBuzz shapes each run, and the runs are laid out in visual order. SkiaFonts.Register(path) adds a font file for Devanagari, Thai, Chinese and other scripts; the MCP server takes them from MATPLOTLIBNET_FONTS. The same shaping pass produces the text measurement, the SVG glyph outlines and the PNG pixels.

3-D charts — 12 series types: Surface, Scatter3D, Bar3D, PlanarBar3D, Line3D, Trisurf3D (Delaunay), Contour3D (marching squares), Quiver3D (vector field), Voxels (face-culled cubes), Text3D (annotations). The 3-D support includes a full Projection3D pipeline, a DepthQueue3D painter's algorithm, shading from Vec3.FaceNormal plus the Color.Shade()/Color.Modulate() extension methods, Svg3DRotationScript for client-side rotation with depth re-sorting, camera-derived back faces (CubeFaceSelection: the panes, cube edges, wall grids and tick rows follow the camera at any azimuth, both server-side and during a drag), a configurable Pane3DConfig (floor/wall colors), and 3D colorbar support.

Streaming & Realtime — StreamingLineSeries, StreamingScatterSeries, StreamingSignalSeries and StreamingCandlestickSeries are backed by RingBuffer<T> and add points with AppendPoint(x, y). A RingBuffer<T> is a fixed-capacity circular sequence of any type (64 M appends/s, and it never allocates on append). StreamingFigure provides throttled re-rendering and auto-scaling axes (SlidingWindow, StickyRight, AutoScale). 11 streaming indicators (SMA, EMA, RSI, Bollinger, MACD, OBV, ATR, Stochastic, WilliamsR, CCI, VWAP) attach automatically to candlestick data. Streaming controls exist for Avalonia, Uno, MAUI, Blazor, and ASP.NET Core. The library includes an SVG diff engine to save bandwidth, and an Rx IObservable<T> adapter.

Geographic projections — the MatPlotLibNet.Geo package has 5 map projections (PlateCarree, Mercator, Robinson, Orthographic, LambertConformal), a GeoJSON parser, embedded Natural Earth 110m data, and GeoPolygonSeries for coastlines, borders and choropleths. A symlog axis scale handles data that spans positive and negative ranges.

Bidirectional SignalR — server-authoritative interactive charts with mutation events (zoom, pan, reset, legend toggle) and notification events (brush-select, hover, the clicked data point). The events form a hierarchy of stacked records, they merge on their own, and the server sends each caller its own hover response.

Accessibility — figure.ToDataTables() gives every chart a data table beside the picture (ToHtml(), ToMarkdown(), ToCsv()). That table is the text alternative that WCAG 1.1.1 requires for a complex image, and an <svg role="img"> cannot provide it on its own. The table is served by MapChartTableEndpoint, by MplChart.ShowDataTable, by the GraphQL chartDataTable field and by the MCP chart_data_table tool, which returns it as markdown to read and as structured values to compute with. The library also has ARIA roles and titles, keyboard navigation, the Okabe-Ito colour-blind-safe palette, the three Petroff accessible colour sequences and a high-contrast theme.

148 colormaps — viridis, plasma, turbo, coolwarm, and 144 more (74 base maps, each with an auto-registered reversed _r variant). The library also has NumPy-style SIMD numerics (Vec, Mat, Linalg, Fft), themes that look like matplotlib's, DataFrame integration with 58 technical indicators, broken axes, and publication-quality SVG/PNG/PDF/GIF export.

License

MIT — free for any use, open-source or commercial, with no copyleft conditions.