LookHere is a lightweight .NET command-line utility that captures any Windows application window as a compact grayscale JPEG screenshot. I built it specifically so that Claude, working through the AI-Bridge MCP server, could actually see what's happening on my screen--something that turns out to be surprisingly tricky to do well within the constraints of MCP tool results.

The Problem

When Claude needs to see a window on my desktop, the screenshot has to travel through the MCP (Model Context Protocol) as a base64-encoded string inside a JSON response. Full-color, full-resolution PNGs can easily run to several megabytes--far too large to fit comfortably in an MCP tool result. And even if you could squeeze them through, Claude doesn't need all that color information just to read a dialog box or check a layout.

LookHere solves this by producing screenshots that are tiny but perfectly readable: grayscale, tone-quantized, and aggressively JPEG-compressed. A typical capture of a desktop window comes in around 50-60KB--small enough to base64-encode and pass through MCP without breaking a sweat.

Download the Setup file here.

How It Works

Under the hood, LookHere does four things:

1. Finds the window. It enumerates all visible windows using the Win32 EnumWindows API and matches against the title substring you provide. No need to specify an exact title--just enough to be unique.
2. Captures it. It uses the Win32 PrintWindow API with the PW_RENDERFULLCONTENT flag, which captures the window's contents even if it's partially obscured by other windows. If PrintWindow fails (some apps don't cooperate), it falls back to a screen-coordinate copy.
3. Converts to grayscale. A ColorMatrix transformation converts the capture to grayscale using the standard luminance weights (0.299R + 0.587G + 0.114B). This alone cuts the information content dramatically.
4. Quantizes tones. Instead of 256 shades of gray, LookHere reduces the palette to a configurable number of tones (128 by default). Fewer distinct gray values means the JPEG compressor has an easier job, producing much smaller files without noticeably affecting readability.

The result gets encoded as a JPEG at a configurable quality level (default 30, which is very aggressive but still perfectly legible for UI elements and text).

Usage

From the command line:

LookHere app="Window Title" [quality=30] [tones=128] [out=path.jpg]

Parameters

Output

LookHere writes the captured image to the output path and prints two lines to stdout:

1. The full path to the output file
2. A summary line like: 1008x793, 128 tones, quality 30, 54KB

Exit codes: 0 for success, 1 for missing arguments, 2 for window not found, 3 for zero-size window (typically minimized).

Built With

LookHere is a single-file .NET 10.0 console application targeting net10.0-windows. It uses Windows Forms (for System.Drawing) but has no external NuGet dependencies. The entire program is under 170 lines of C#.

Win32 APIs used:

• EnumWindows / GetWindowText -- window enumeration
• PrintWindow -- off-screen window capture
• GetWindowRect -- window dimensions

Why Grayscale?

You might wonder why I didn't just compress harder or resize the image. Resizing throws away spatial detail--text becomes unreadable. Heavy JPEG compression on color images creates ugly chroma artifacts. But grayscale sidesteps both problems: there's no color information to smear, and the compressor only has one channel to worry about. The result is a file that's a fraction of the size of a color JPEG at the same resolution, with text that's still sharp and legible.

The tone quantization is the other secret weapon. By snapping 256 possible gray values down to 128 (or fewer), you create large regions of identical pixels that JPEG's DCT compression handles beautifully. It's like giving the compressor a head start.