The agentfenster MCP server

Last updated: August 8, 2026

The agentfenster MCP server lets Claude Code and other MCP clients drive real Windows applications on a hidden second desktop inside your own session, with the same logins, cookies and VPN, while you keep using your machine. It is a local stdio server that ships with agentfenster, a paid Windows app, and exposes eleven tools: from a one-call task loop to single verified clicks.

Every action is verified against the window state afterwards. A click that lands nowhere is reported as a failure, never as a silent success.

Install

Install agentfenster on Windows first; the MCP server is part of the app and every plan starts with a 3-day trial. Then register the server with Claude Code:

claude mcp add agentfenster -- agentfenster-mcp

Then just ask. For example:

> open the character map on the agent desktop and turn on the advanced view

How it works

The agent works on a second Win32 desktop inside your session. It never takes your foreground, mouse or keyboard, and you can watch it live in the app.
Tools read the UI Automation element tree instead of screenshots: one line per interactive element, capped at 200 elements per window.
The click loop of desktop_task runs inside the server on a model endpoint you configure, so its retries and intermediate steps never enter your context.
Risky actions such as delete, send, pay or install are denied by default on MCP calls and reported, instead of blocking on a confirmation nobody can answer.

The eleven tools

Tool	What it does
`list_windows`	Lists the windows this machine can drive, with title, class, minimized state and whether a picture can be captured. The starting point when you do not know the exact window title.
`desktop_task`	Runs a full GUI task in one window and returns a single paragraph: what happened, whether it worked, and if not, which step failed and why. The click loop runs inside the server, so its retries never enter your context.
`inspect_window`	Returns the capped, filtered element tree of one window as a compact list, for planning clicks yourself. Capped at 200 elements and never walked from the desktop root.
`replay`	Lists recent runs, or returns one run's step-by-step log: every plan, every action, whether it changed anything, and a link to the visual replay in the app.
`open_app`	Launches a program on the shared agent desktop and reports its first titled window. The agent desktop starts empty, so this is how anything gets onto it.
`read_desktop`	Reads the compact element tree of a window on the agent desktop, one line per interactive element, no screenshot. Plan your clicks from this.
`click`	Clicks one element, chosen by id or name from read_desktop, then verifies against the desktop state and states explicitly whether anything changed.
`type_text`	Types text into one element on the agent desktop, then verifies whether anything changed.
`press_key`	Presses a key or key combination such as ctrl+s, enter, tab or alt+f4, optionally focusing an element first, with verification afterwards.
`scroll`	Scrolls a window or a specific pane and reports whether the content actually moved. A list that is already at the bottom reports no effect.
`agent_windows`	Lists the windows currently open on the shared agent desktop, frontmost first, so you can pick one by a title substring.

Frequently asked questions

Do I need the agentfenster app to use the MCP server?

Yes. The MCP server ships with agentfenster, a paid Windows app. Every plan starts with a 3-day trial, no card and no account required.

Which MCP clients can connect?

Any client that can launch a local stdio MCP server. With Claude Code, run: claude mcp add agentfenster -- agentfenster-mcp

Does the agent take over my mouse or screen?

No. The agent works on a hidden second desktop inside your own Windows session, with the same logins, cookies and VPN. You keep using your machine and can watch the agent desktop live in the app.

What happens when a click has no effect?

It is reported as a failure, never as a success. Every action is verified against the window state afterwards, and every tool states explicitly whether anything changed.

Does the server send data anywhere?

agentfenster sends no telemetry. The only network traffic is the model endpoint you configure for the click loop.

Questions

Join the Discord to watch the agent work live and get help, or read the Privacy Policy for how agentfenster handles data.