DesktopDriver
StockSharp DesktopDriver
Drive desktop applications built on Avalonia, WPF or MAUI from tests, from a command line, or from an AI agent over MCP.
DesktopDriver reads the interface, not the screen. A table answers with its columns and rows, a tree with its items, a workspace with its panels - in words that do not change with a theme, a font or a language. Input goes through the toolkit's own pipeline, and only into a run that cannot reach anything real. Pictures are there for a person to look at; tests check what the controls say.
How it works
An application that can be driven references the bootstrap of its toolkit. Started the ordinary way it opens nothing. Started with
--ui-automation, it opens a local named-pipe endpoint restricted to the current OS user, and writes where the endpoint is to the file named by--ui-automation-endpoint=<path>.A runner - a test, the command line, the MCP server - starts the application that way (or attaches to one somebody else started), checks the product, instance and protocol in the session handshake and asks: which windows are open, what is in this one, what does this table show. It can then click, type, press keys and scroll, and wait for the answer to change.
Adapters turn controls into what the protocol knows: grids, trees, lists, pickers, workspaces, charts, documents, diagrams. An application registers adapters for its own controls through the same interface the standard ones use.
Related MCP server: Smooth Operator Agent Tools
NuGet packages
Install the CLI or MCP server with the .NET 10 SDK:
dotnet tool install StockSharp.DesktopDriver.Cli --tool-path tools
dotnet tool install StockSharp.DesktopDriver.Mcp --tool-path toolsThe commands are tools/desktop-driver and tools/desktop-driver-mcp. To add automation to an
application, reference the bootstrap package for its toolkit. To drive it from code, use the client,
runner and testing packages.
For MCP Registry publication and Claude Desktop extensions, see MCP distribution.
Package | What it is |
The protocol: requests, answers, states, addresses | |
Reading, identity, revisions and waiting, independent of any toolkit | |
The endpoint inside the application, and | |
The typed client every runner talks through | |
Starts, finds and reaches applications; the catalogue of what may be started | |
What an MSTest suite needs on top of the runner | |
The command line | |
The MCP server, see its README | |
Avalonia: reading, input and pictures of the standard controls | |
Avalonia: the | |
Avalonia: a Dock.Avalonia workspace, read as panels and groups | |
Avalonia: input for headless tests | |
Avalonia: one call that makes an application drivable | |
WPF: reading, input and pictures of the standard controls | |
WPF: one call that makes an application drivable | |
MAUI: reading, input and pictures of the standard controls | |
MAUI on Windows: one call that makes an application drivable |
Make an application drivable
Avalonia, once the main window exists (the sample does the same with the grid module alone):
public override void OnFrameworkInitializationCompleted()
{
if (ApplicationLifetime is IClassicDesktopStyleApplicationLifetime desktop)
{
desktop.MainWindow = new MainWindow();
if (UiAutomationLaunch.TryRead(desktop.Args) is { } launch)
{
_automation = AvaloniaAutomationBootstrap.Start(
UiAutomationStartup.For("my.app", version, fixtureId: null, hasLiveOutsideWorld: false, isTestProfile: false),
launch.EndpointFile,
[binder => new DataGridAutomationModule(binder), binder => new DockAutomationModule(binder)]);
}
}
base.OnFrameworkInitializationCompleted();
}Dispose the returned runtime when the application exits. WpfAutomationBootstrap and
MauiAutomationBootstrap take the same arguments.
The sample applications show startup and shutdown for all three toolkits. The WPF and MAUI Windows samples also have tests that start their executables, connect over the local pipe, read their controls, send input and retrieve screenshots.
UiAutomationStartup.For decides whether input is allowed. An application that talks to nothing outside
its own process is always safe to click through. One that connects to something real - a broker, a
store, a service - is safe only when it was started on a test profile that replaces it; otherwise the
endpoint still reads, but every input is refused.
Many applications ship two builds: an ordinary one with no driver in it, and one made to be driven. The driver is only referenced by the second, so the first has nothing listening by construction.
Drive it
From a test:
await using var app = await DrivenApplication.StartAsync(executable, extraArguments: null, TimeSpan.FromSeconds(60), cancellationToken);
await using var client = await app.ConnectAsync("my.app", cancellationToken);
var windows = await client.GetSurfacesAsync(cancellationToken);From a command line, against an application somebody started with --ui-automation:
tools/desktop-driver find --endpoint endpoint.json --kind grid
tools/desktop-driver grid-rows --endpoint endpoint.json --node window:MainWindow/Instruments --limit 5desktop-driver with no arguments prints every operation.
From an agent: register the MCP server, see DesktopDriver.Mcp.
Its MCP name is StockSharp.DesktopDriver, and its display title is StockSharp DesktopDriver.
Documents
What answers for which control, and how to add an adapter
Build and test
dotnet build DesktopDriver.slnx
dotnet test DesktopDriver.Runtime.Tests --filter "FullyQualifiedName~Revision"The MAUI projects need the MAUI Windows workload. MCP application tests, WPF screenshot tests and
DesktopDriver.Windows.Tests need an interactive Windows desktop session. Run the filtered suites
with hang and session timeouts using ./.github/scripts/Test.ps1 -Suite Headless or -Suite Desktop
after a Release build.
Packages from source
dotnet pack DesktopDriver.slnx -c Release -o artifacts/packagesSources instead of packages
A product checked out beside this repository can reference the projects themselves:
<ProjectReference Include="..\DesktopDriver.Mcp\DesktopDriver.Bootstrap.Avalonia\DesktopDriver.Bootstrap.Avalonia.csproj" />A product that builds a second, drivable copy of itself by passing a property on the command line adds
GlobalPropertiesToRemove="<that property>" to the reference: the driver has no build of its own for it, and
would otherwise be built twice into the same folder.
License
See LICENSE.
This server cannot be deployed
Related MCP Connectors
remote debug iOS/Android/Unity/Godot/Flutter/RN/Web on real-device.ui-tree/screenshots/taps,tests.
MCP server for visual regression testing: triage a PR's UI diffs from your coding agent.
Drive and test your React Native and Expo apps end-to-end on iOS and Android. Supports real devices, simulators, and emulators. Snapshots, taps, typing, swipes, screenshots, deep links, reloads, builds and runtime errors.
Drive real devices from your AI Coding tool. Embed a client SDK (Unity, Godot, Flutter, iOS/macOS, Android, React Native, Web) in your app, then capture screenshots, traverse the UI tree, inject taps and key events, and run automated test tasks on the physical device over a secure relay.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceProvides automated GUI testing and control capabilities through an MCP server that enables mouse movements, keyboard input, screen captures, and image recognition across Windows, macOS, and Linux.46MIT
- AlicenseNot gradedqualityDmaintenanceWindows automation MCP offering * AI Vision (e.g. Click by Description) * Windows UI Automation Tree tools * Chrome Automation via Playwright * Mouse control * Keyboard control * a lot more (>40 tools) Also comes with Python/TypeScript/C# client libs and a Windows Desktop tool to try all the tools.2MIT
- AlicenseNot gradedqualityCmaintenanceMCP server for AI agents to automate Windows WPF apps with semantic UI Automation selectors, screenshots, recording, and FlaUI test generation.2MIT
- AlicenseNot gradedqualityCmaintenanceA Model Context Protocol server that gives AI assistants full control over native Windows applications — launch, inspect, click, type, screenshot, and test any WinUI3, WPF, WinForms, UWP, or Win32 app.72 npm42MIT