Control Any Windows App with AI — No API Required
Let Claude, Cursor, VS Code or ChatGPT click buttons, fill forms, and read the interface of any native Windows app through UI Automation — even apps with no API at all. Free, local, no admin rights needed.
To let Claude click buttons and fill forms in any Windows app, install LMCP — a free local MCP server that drives the Windows UI Automation framework, the same accessibility layer screen readers use. Because it reads and controls the interface directly, it works with any native Windows app, even ones with no API or automation support at all.
What You Can Do
- Read the UI tree — get a structured, labeled map of a running app's buttons, fields, and menus
- Find and wait for elements — locate a specific control, or wait for one to appear before acting
- Click — by element or by screen coordinate
- Type text — into the focused control, or a specific one
- Send key combinations — like
ctrl+sorctrl+shift+n - Click a system tray icon — by its label, without hunting for pixel coordinates
This is a different mechanism from LMCP's Outlook, Teams, or OneDrive tools — those talk to an app's own data or COM interface. UI Automation instead drives the interface itself, which is what makes it work with apps that expose nothing else.
How to Install
Download LMCP for Windows and install it:
- Run the downloaded
LMCP-Setup.exefrom your Downloads folder - Follow the installer — no admin rights needed, it installs to your user folder
- LMCP starts automatically and appears in your system tray
Takes about 30 seconds. Your AI clients are configured automatically.
Then restart your AI client and try it against any open app.
Example Prompts
- “Read the UI of the app I have open and tell me what controls are visible” — maps the interface first, before acting
- “Can Claude click buttons and fill forms in any Windows app for me?” — yes, this is exactly what UI Automation is for
- “Automate a legacy desktop app with AI without an API” — find the fields, type into them, click submit
- “Wait for the export to finish, then click Save” — waits for an element to appear before acting
Real-World Workflow
Repetitive data entry in a legacy app
Read the UI tree of this legacy inventory app, find the “New Entry” form, and fill it in with the values from this spreadsheet, one row at a time.
Your AI maps the form first, then drives it — no macro recorder, no screen-coordinate guessing that breaks the moment a window resizes.
LMCP also connects your AI assistant to window and monitor management, Outlook and Microsoft Teams on Windows. See the full list of guides or learn more at local-mcp.com.
Limitations
- Windows-only — UI Automation is a Windows accessibility framework; there is no equivalent tool on macOS or Linux in LMCP.
- Reads and drives what's on screen — a minimized or hidden window may not expose all elements the way a foreground one does.
- Layout-dependent — an app that changes its interface between versions can change which elements LMCP finds; re-reading the UI tree after an update is the fix.
- Not a replacement for a native integration — for apps LMCP already has a dedicated tool for (Outlook, Teams, OneDrive), those are faster and more reliable than driving the UI.
Frequently Asked Questions
Can Claude click buttons and fill forms in any Windows app for me?
Yes — LMCP drives the Windows UI Automation framework, which works with any native Windows app, including ones with no API at all.
Can I automate a legacy Windows desktop app with AI without an API?
Yes. Since UI Automation reads and controls the interface directly, it does not need the app to expose any automation surface of its own.
Does this need admin rights?
No — LMCP installs and runs per-user. Some apps that run elevated may still require LMCP to run elevated to control them; that is a Windows permission model detail, not an LMCP requirement.
Related Guides
Comments
Questions and experiences welcome — comments are reviewed before they appear.