Skillsdsh-plugin
mcp-ds-ocr
dsh-ocr is a DeepSeek Harness plugin that provides OCR visual recognition via OpenAI or Anthropic vision model APIs. It adds an ocr_recognize tool, configurable in DSH settings with model list and test connection buttons, returns percent/pixel coordinates with confidence, and re-reads low-confidence regions automatically.
Stars
1
Forks
0
Open issues
0
Last push
Aug 16, 2026
Latest release
—
24h Growth
+0
7d Growth
+0
Installation
No reliable install command was detected. Check the project README for installation instructions.
Overview
dsh-ocr is a DeepSeek Harness plugin that provides OCR visual recognition via OpenAI or Anthropic vision model APIs. It adds an ocr_recognize tool, configurable in DSH settings with model list and test connection buttons, returns percent/pixel coordinates with confidence, and re-reads low-confidence regions automatically.
- OCR recognition via OpenAI-compatible or Anthropic vision APIs, selectable apiFormat.
- Configurable in DSH settings: API URL, API key, model, confidence threshold, coordinate mode, and retry options.
- Get model list and test connection buttons in the plugin settings card.
- ocr_recognize tool supports absolute or workspace-relative image paths for PNG/JPEG/GIF/WebP up to 10 MB.
- Returns text with pixel and percentage bounding boxes, diagonal two-point description, confidence, and retried flag.
- Low-confidence regions are automatically re-read and merged using IoU matching.
- Optional manual focus regions and prompt hint for targeted recognition.
- Settings persisted via dsh-settings; HTTP calls made with curl through local child process.