Grok Image

io.github.codeChap由 GitHub 账号 codeChap 发布v0.1.0更新于 Oct 24, 2026

MCP server for Grok image generation and editing

发布方自己的服务:Surface

已验证STDIO仅桌面媒体与设计AI 与机器学习

概览

AI 生成的概览

一个本地 MCP 服务器,让助手通过 xAI 的 Grok 图像 API 生成和编辑图片,并提供企业头像处理工具。

可以做什么
根据文字提示生成图片,可设置模型、数量、宽高比、分辨率、质量和风格,并返回实际发送给 API 的提示词。用自然语言指令编辑一张或最多五张现有图片,多图时按索引引用。从人像生成加白边补齐的企业头像,可指定服装、需保留细节、代词和尺寸参数。列出内置与自定义图像风格及其提示词模板。
适用场景
适合希望助手在对话中直接创建或修改图片、而不必切换到独立图像工具的场景,例如绘制插图、给照片换风格,或把肖像整理成标准头像。
运行要求
以本地进程方式运行,需用 Rust(2024 edition)从源码构建,未发布安装包。需要 console.x.ai 提供的 xAI API 密钥,写入 ~/.config/mcp-server-grok-image/config.toml 配置文件,并需能访问 Grok 图像 API 的网络。在客户端配置中以构建出的可执行文件路径注册。
安装前请注意
API 密钥以明文保存在配置文件中,需妥善保护。生成和编辑图片会调用付费的第三方 API,可能按图片产生费用。源图片(包括本地文件路径)会被发送到 xAI。头像工具会向保存目录写入中间文件和输出图片。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Grok Image,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

mcp-server-grok-image

An MCP (Model Context Protocol) server for xAI's Grok image generation API. Built in Rust, exposes image generation and editing as MCP tools.

Communicates via stdio using JSON-RPC 2.0, like all MCP servers.

Tools

ToolDescription
generate_imageGenerate an image from a text prompt
edit_imageEdit an existing image using natural language instructions
headshotCorporate headshot from a source portrait (pad to 3:2 + fixed edit prompt)
list_stylesList available image styles for use with generate_image

generate_image

Generate an image from a text description.

Parameters:

NameTypeRequiredDescription
promptstringyesText description of the desired image
modelstringnoModel to use (default: grok-imagine-image-2.0)
nintegernoNumber of images to generate (1-10, default 1)
aspect_ratiostringnoAspect ratio: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 2:1, 1:2, 19.5:9, 9:19.5, 20:9, 9:20, 21:9, 5:2, auto
resolutionstringnoOutput resolution: 1k (~1024px, default) or 2k (~2048px)
qualitystringnolow, medium, or auto (2.0 only; omitted = auto. Auto currently serves low for generation)
response_formatstringnoOutput format: url (default, temporary) or b64_json
stylestringnoStyle name to apply (use list_styles to see options)

When a style is set, the prompt is wrapped in the style's template. For example, with style: "watercolor" and prompt: "a cat on a roof", the API receives "a cat on a roof, as a watercolor painting". Avoid including style language in the prompt itself when using this parameter.

The response includes the resolved prompt so you can see exactly what was sent to the API.

edit_image

Edit an existing image using natural language instructions.

Parameters:

NameTypeRequiredDescription
image_urlstringno*URL, base64 data URI, or local file path of the source image. Mutually exclusive with images.
imagesstring[]no*Up to 5 source images for multi-image editing. Reference them in the prompt as <IMAGE_0>, <IMAGE_1>, …
promptstringyesNatural language edit instructions
modelstringnoModel to use (default: grok-imagine-image-2.0)
nintegernoNumber of variations to generate (1-10, default 1)
aspect_ratiostringnoSame set as generate_image, including 21:9 and 5:2
resolutionstringnoOutput resolution: 1k (~1024px, default) or 2k (~2048px)
qualitystringnolow, medium, or auto (2.0 only; omitted = auto. Auto currently serves medium for editing)
response_formatstringnoOutput format: url (default, temporary) or b64_json

* Provide either image_url or images.

Note: The style parameter is intentionally not available on edit_image -- edit prompts are instructions (e.g. "remove the background"), not descriptions, so wrapping them in style templates would produce nonsense.

headshot

Expand-only portrait fix (Gemini pipeline equivalent on Imagine). Does not reframe pose, cut out hair, or redesign the person.

  1. Resize full source (default 550px wide) — never crop
  2. Letterbox with white gutters to canvas width (default 780)
  3. Call grok-imagine-image-2.0 at quality medium: complete cut-off shoulders if needed; clean solid white background; keep face/hair/pose/clothing/logos

No cutout / no rembg / no transparent alpha — same job as the original Gemini headshot skill.

Parameters:

NameTypeRequiredDescription
imagestringyesLocal path, http(s) URL, or data: URI
clothingstringnoFor missing-shoulder fill only
notesstringnoMust-preserve details (glasses, exact logo text, …)
pronounstringnohis / her / their (default their)
gravitystringnoLetterbox gravity (North default)
content_widthintegernoResize width before pad (default 550)
canvas_widthintegernoPadded width (default 780)
resolutionstringno1k or 2k (default 2k)
output_pathstringnoOptional final path (also under save_dir)
nintegernoVariations (1–10, default 1)
qualitystringnolow / medium / auto (default medium)
modelstringnoDefault grok-imagine-image-2.0

Padded intermediate: save_dir/headshot-padded_*.jpg.

list_styles

Returns all available image styles with their name, description, and prompt template. No parameters.

Built-in Styles

StyleDescription
watercolorWatercolor painting style
oil-paintingOil painting with visible brushstrokes
pencil-sketchDetailed pencil sketch
pixel-artRetro pixel art
animeAnime style illustration
pop-artBold pop art style
art-nouveauArt nouveau with flowing organic lines
cinematicCinematic photography with dramatic lighting
portraitProfessional portrait photography
macroExtreme macro photography
aerialAerial drone photography
studioStudio photography on clean background
noirDark film noir style
vintageFaded vintage photograph

Available Models

ModelNotes
grok-imagine-image-2.0 (default)Optional quality (low / medium / auto), up to 5 edit references, 21:9 and 5:2. Auto currently serves low for generation and medium for editing.
grok-imagine-image1.0. Still available; no quality param.
grok-imagine-image-qualityRetires 2026-11-02. After that the slug is served by grok-imagine-image-2.0 at quality: low ($0.01 less per image than the quality model).

Prerequisites

Setup

Create the config file:

bash
mkdir -p ~/.config/mcp-server-grok-image

Create ~/.config/mcp-server-grok-image/config.toml:

toml
api_key = "xai-..."

Custom Styles

Add custom styles to your config file. Custom styles with the same name as a built-in will override it.

toml
api_key = "xai-..."
[[styles]]name = "my-style"description = "My custom look"template = "{prompt}, in my custom style"
[[styles]]name = "watercolor"description = "My watercolor variant"template = "{prompt}, as a loose expressive watercolor with ink outlines"

Templates must contain the {prompt} placeholder. Any custom style missing it will be skipped with a warning at startup.

Build

bash
cargo build --release

This produces target/release/mcp-server-grok-image.

For development:

bash
cargo build              # debug buildcargo run                # run in dev modeRUST_LOG=debug cargo run # run with debug logging

MCP Configuration

Add to your Claude Desktop config (~/.config/Claude/claude_desktop_config.json):

json
{  "mcpServers": {    "grok-image": {      "command": "/path/to/mcp-server-grok-image"    }  }}

Project Structure

src/  main.rs      process entry (stdio MCP)  config.rs    TOML / env config  styles.rs    built-in + custom styles  grok.rs      xAI request/response types  params.rs    MCP tool params + validation  image_io.rs  data URIs, local files, mime, fetch  headshot.rs  letterbox pad + expand prompt  server.rs    MCP tools and Grok HTTP

License

MIT

来源:README.md,提交 6382b52

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.1.0最新Sep 16, 2026

评分

已满足适用检查项的 40%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。

其他 Grok 集成