opencode-clipboard-ocr

Paste screenshot. Extract text locally. Send OCR to any OpenCode model.

opencode-clipboard-ocr lets OpenCode read text from clipboard screenshots without switching to a vision model. It uses Windows OCR locally, saves the screenshot for debugging, and sends only extracted text to your current model.

What It Is

• Clipboard screenshot OCR for OpenCode.
• Good for image-based errors, terminal output, logs, and UI text.
• Works with text-only models like DeepSeek, Qwen, Llama, and local models.
• Does not send the image to the model.
• Does not require a vision model.

Limits

• Windows-only for now.
• OCR is not visual understanding. It reads text, not diagrams/layout.
• OCR quality depends on screenshot clarity.
• OpenCode may briefly show a blank waiting state while the command hook runs OCR.

Install

Clone this repo, then run:

powershell -ExecutionPolicy Bypass -File .\install.ps1

Restart OpenCode.

Usage

1. Copy a screenshot/image to the Windows clipboard.
2. In OpenCode, run:

/paste-image what error is shown?

The model receives:

Image Content:

<OCR text>

Your current OpenCode model stays active.

Files

The plugin saves source screenshots under each project:

.opencode/clipboard-images/

Add this to your project .gitignore:

.opencode/clipboard-images/

Temporary OCR variants are deleted after extraction.

Uninstall

powershell -ExecutionPolicy Bypass -File .\uninstall.ps1

How It Works

OpenCode loads local/global plugins from:

~/.config/opencode/plugins/

This installer copies:

plugins/paste-image.js -> ~/.config/opencode/plugins/paste-image.js
commands/paste-image.md -> ~/.config/opencode/commands/paste-image.md

The command hook reads the Windows clipboard image, saves it as PNG, runs Windows OCR, preprocesses terminal screenshots with a few high-contrast variants, and replaces the command prompt with OCR text.