visionpower
@piercehuang663
VisionPower: a portable image understanding MCP server for Codex, Claude Desktop, Cursor, and other agents.
Image recognition for text-only DeepSeek Harness models: registers the analyzeimage tool, routing images to a dedicated vision model (default zhipu/glm-4.6v-flash) through the harness's own ctx.llm seam.
Description machine translated from the original.
npx @deepseek-ai/dsh plugin add dsh-plugin-vision-tool