dsh-opencode-eyes-nvidia

DeepSeek Harness plugin: eyes for text-only models through NVIDIA NIM. It mounts the bundled opencode-eyes-nvidia MCP server, which sends an image to a hosted vision model (MiniMax-M3 by default) and returns a description, exposing mcp__opencode_eyes_nvidia__describe_image and mcp__opencode_eyes_nvidia__list_vision_models. 适合需要利用英伟达托管的视觉模型来识别和描述图片的用户。

Package

dsh-opencode-eyes-nvidia

Compatibility

Unverified

Harness peer range

>=0.1.5-rc.1 <0.2.0-0 || >=0.2.0-rc.0 <0.3.0-0

Cordis peer range

>=4.0.2

Version

0.1.0

License

MIT

Last updated

Oct 8, 2026

Configuration

Key	Environment variable	Default	Meaning
`python`	—	discovered	interpreter that runs the server
`apiKey`	`NVIDIA_API_KEY`	(empty)	the first key of the rotation ring
`model`	`NVIDIA_MODEL`	`minimaxai/minimax-m3`	default model id
`baseUrl`	`NVIDIA_BASE_URL`	`https://integrate.api.nvidia.com/v1`	NIM endpoint
`timeoutSeconds`	`NVIDIA_TIMEOUT`	`120`	the server's own HTTP timeout
`maxDimension`	`NVIDIA_MAX_DIMENSION`	`2048`	images are downscaled to this edge length
`jpegQuality`	`NVIDIA_JPEG_QUALITY`	`85`	JPEG quality of the re-encoded image
`thinkingMode`	`NVIDIA_THINKING_MODE`	(unset)	`enabled` / `disabled` / `adaptive`; only sent to `minimax` models
`toolCallTimeoutMs`	—	`300000`	DSH's per-call budget; key rotation sleeps between attempts
`env`	—	`{}`	raw passthrough — use this for `NVIDIA_API_KEYS` and `NVIDIA_API_KEY_1..N`

Every field is optional and lives in the loader row. For example, in cordis.patch.yml:

- id: dsh-opencode-eyes-nvidia
  name: 'dsh-opencode-eyes-nvidia'
  config:
    apiKey: 'nvapi-...'
    model: 'minimaxai/minimax-m3'
    env:
      NVIDIA_API_KEYS: 'nvapi-first,nvapi-second'

dsh-opencode-eyes-nvidia

Install

README

Configuration