FlaUI-MCP
An MCP server for Windows: an agent drives windows through the accessibility tree and element refs, no screenshot coordinates.
- 113 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent controls any window on the computer: clicks and input run as your account.
Install
- Works in
Manual install
Download the archive for your machine (`win-x64` or `win-arm64`) from the releases page, unpack it and add a command with the path to `FlaUI-MCP.exe` to the MCP configuration. Call `windows_list_windows` first, then `windows_snapshot` for the target window.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent controls any window on the computer: clicks and input run as your account.
This is third-party code. Review the repository files before installing.
What it does
FlaUI-MCP brings the Playwright approach to Windows desktop apps. The `windows_snapshot` tool returns an accessibility tree with short refs such as `w1e5`, and `windows_click`, `windows_type`, `windows_fill` and `windows_send_keys` act on an element by ref. It launches and closes programs, lists windows, sets focus, reads text and saves a window as PNG, including in the background. `windows_batch` runs several actions in one call. Each call has a 30-second limit so a stuck dialog cannot hang the server. It runs on UI Automation through FlaUI.
Who it is for. Developers and testers of Windows desktop apps who want an agent to press buttons and read fields by semantics instead of pixels.
Good fit when
- You need to automate an old program that has no API and no web interface.
- A WPF or WinForms UI test should be described through element refs.
- An agent walks a scenario in Calculator or Notepad and checks the result.
- A series of clicks is packed into one batch call for speed.
Not a fit when
- You need browser automation: Playwright MCP fits better.
- The target system runs macOS or Linux.
- The application draws its own UI and exposes no accessibility tree.
Example request
Launch Calculator, compute 3 × 3 through the accessibility tree and report the value on screen.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The server works only on Windows 10 and 11 and drives live windows on your computer, so the agent can press any button, including destructive ones. The tree's completeness depends on the app; custom elements without accessibility stay invisible. Each call is cut off after 30 seconds. The build without the self-contained bundle needs .NET 8. A background screenshot falls back to the normal path when the frame is blank. The latest commit is dated July 2026.
How to disable. Remove the server entry from the MCP client configuration and the folder with the unpacked archive.
FAQ
What is an element ref?
A short identifier from the accessibility tree, such as `w1e43`. Buttons and fields are addressed by it, with no screen coordinates.
Which archive should I pick?
`win-x64` for 64-bit Windows, `win-arm64` for ARM. The `self-contained` variant needs no installed .NET 8.
Can several actions run at once?
Yes, `windows_batch` takes a list of actions (click, type, snapshot) and returns a result per step.
How do I save a window capture?
`windows_screenshot` with `savePath` set to an absolute `.png` path. An existing file is replaced only with `overwrite: true`.
What happens with a frozen window?
After 30 seconds the call returns a clear error and the server keeps running.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server and wincli CLI: agents drive Windows apps by control name through UI Automation instead of screen coordinates
An MCP server for driving Windows: the agent opens apps, clicks and types, reads window and UI state, and works with files and PowerShell
The agent moves the mouse, types, drives windows and captures PNG screenshots of a Windows desktop through AutoIt.
DesktopManager gives an agent Windows windows and monitors over MCP: it starts read-only and changes need separate flags.
A Swift MCP server that lets an agent drive any macOS app: click, type and read the interface tree
A macOS CLI and MCP server: screenshots, accessibility-based UI inspection and app control via clicks and typing
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| FlaUI-MCP (this card) | no | not required | 113 | |
| Windows-MCP | no | not required | 11,574 | drives Windows through screenshots and general actions; FlaUI-MCP uses the UI Automation tree and element refs |
| MCP Windows Desktop Automation | no | not required | 119 | another Windows desktop server; FlaUI-MCP mirrors Playwright's snapshot and ref scheme |
README in short
The README explains the idea (Playwright for Windows), shows a Calculator example, lists tools and archive variants, and gives client configuration.