image-gen-mcp

image-gen-mcp

**Configurable, provider-agnostic image generation, as a DeepSeek Harness

plugin.** Point it at any OpenAI-compatible /v1/images/generations endpoint

and the model gains image_generate and image_config_status — plus a bundled

skill that tells it when to use them.

*A Codex plugin that exposes image generation tools through MCP with a

configurable base URL and API key.*

Install

DeepSeek Harness Desktop — open Plugins in the sidebar, choose **Add

plugin**, and enter:

https://github.com/bauerelizabeth07139/image-gen-mcp

Then switch the new dsh-image-gen-mcp bundle on, and give it a backend (see

[Configuration](#configuration) — without a base URL and key the plugin

deliberately mounts nothing and says so).

dsh CLI — install it into the profile you actually boot:

dsh plugin --profile web add bauerelizabeth07139/image-gen-mcp

No git on the machine? pnpm resolves a git shorthand with git ls-remote,

which fails with 'git' is not recognized when git is missing. Use the tarball

instead — that path is plain HTTPS:

dsh plugin --profile web add https://codeload.github.com/bauerelizabeth07139/image-gen-mcp/tar.gz/master

The same address works in the Desktop Add plugin dialog. Replace master

with a commit SHA to pin an exact revision (/tar.gz/).

Uninstall with dsh plugin --profile web remove dsh-image-gen-mcp.

Requirements

• Node.js ≥ 22 — the harness itself; the server is plain Node with no npm

dependencies, so there is nothing to install.

• A backend: a base URL that answers POST {baseUrl}/v1/images/generations

and an API key for it. The base URL must not include /v1.

Tools

| Tool | What it does |

|---|---|

| mcp__image_generation__image_generate | Generates an image from a prompt. Parameters: prompt (required), model, size, n. |

| mcp__image_generation__image_config_status | Reports whether the base URL and key are configured — ask this first when a generation fails for configuration reasons. |

Bundled skill

skills/image-generation/SKILL.md is registered by the plugin, so the model

knows the workflow: check image_config_status, generate with

image_generate, return the artifact or an actionable error. It is adapted from

the repository's original Codex skill — same workflow, harness tool names, plus

a note about base64 payloads.

Configuration

| Key | Environment variable | Default | Meaning |

|---|---|---|---|

| baseUrl | IMAGE_GEN_BASE_URL | (none — required) | backend base URL, without /v1 |

| apiKey | IMAGE_GEN_API_KEY | (none — required) | credential for that backend |

| model | IMAGE_GEN_DEFAULT_MODEL | (empty) | default model; the server picks dall-e-3 in openai mode and gpt-image-1 in generic mode when this is empty |

| provider | IMAGE_GEN_PROVIDER | generic | generic (URL response) or openai (base64 response) |

| timeoutMs | IMAGE_GEN_TIMEOUT_MS | 30000 | the server's own HTTP budget |

| toolCallTimeoutMs | — | 120000 | DSH's per-call budget |

| env | — | {} | raw environment passthrough |

Example loader row:

- id: dsh-image-gen-mcp
  name: 'dsh-image-gen-mcp'
  config: