Distribea Mcp

io.github.Distribea由 GitHub 账号 Distribea 发布v2.1.0更新于 Oct 26, 2026

Generate images, video, music and voice from your CLI or AI agent. On-brand AI media toolkit.

已验证Streamable HTTP可网页运行开发者工具媒体与设计

概览

AI 生成的概览

Distribea 让助手用自然语言提示生成符合品牌风格的图片、视频、音乐和配音,并可整页填充或替换站点图片。

可以做什么
  • 生成媒体:根据简短主题生成单张图片、短视频、音乐或配音,可指定模型或参考照片。
  • 填充与替换页面图片:把页面或整个项目中的占位图、图库图替换为品牌风格图片并写回代码,先给出免费方案。
  • 编辑与收尾图片:修图、去背景、放大、扩图,补全缺失的替代文本并把大图转为 WebP。
  • 品牌与目录:锁定站点视觉风格、生成 logo 与 favicon 包、锁定固定人物或产品,并浏览带参考价格的模型目录。
适用场景
适合项目需要真实图片或音频而非图库照片和灰色占位图的场景:落地页、商品图、博客封面、主视觉短片、背景音乐与配音,或在 AI 编程助手中就地改造已有站点。
运行要求
可作为远程端点使用,也可作为本地 npm 包(npx distribea-mcp)运行,需要 Node.js。需要 Distribea 账号与套餐;认证方式为浏览器 OAuth 登录,或通过 DISTRIBEA_MCP_KEY 环境变量或 Authorization 请求头提供个人访问密钥。需要访问托管引擎的网络连接。
安装前请注意
生成按 Distribea 套餐的积分计费,视频、音乐和配音需要付费套餐;可先用试运行选项查看价格。访问密钥(DISTRIBEA_MCP_KEY 或 Authorization)属于凭据,不要写入共享配置。整页填充与替换工具会写入项目文件,但原图会保留为 .original。请求会把你的描述、页面的一小段文本、哈希后的项目标识以及你指定编辑的文件发送到托管引擎及其 AI 服务商。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Distribea Mcp,将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.Distribea%2Fdistribea-mcp/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "distribea": {
      "type": "http",
      "url": "https://distribea.com/api/mcp/remote"
    }
  }
}

README

Distribea

Generate images, video, music & voice — right inside your terminal and your AI agent.

[npm version] [npm downloads] [MCP registry] [License: MIT] [GitHub stars]

CLI · MCP server · Claude Skill — works with Claude Code · Cursor · Windsurf · Replit · Copilot

Distribea remakes any video with you in it (Motion Control) and gives a royalty-free library of music, sound effects, footage and images. This package is its on-brand AI media toolkit for builders: it generates images, video, music and voice from plain-language prompts and ships them production-ready — optimised WebP with SEO alt text, downloaded for you or patched straight into your code. One connector, four media types, reachable three ways: a CLI in your terminal, an MCP server your AI coding agent calls while it builds, and a Claude Skill. Define your brand once; a built-in AI art director writes every prompt — you say what you want, never how to render it. Billed in credits at the regular site price; the connector holds no API keys and no secrets.

bash
npm install -g distribea-mcpdistribea login                                       # opens your browser — no key to paste
distribea image "founder portrait, soft daylight"  --portraitdistribea video "slow waves rolling at golden hour" --landscapedistribea music "warm lo-fi beat, mellow, 90 bpm"distribea voice "Welcome to our store."             --lang en

Each file lands in a tidy folder, ready to use. Add --dry-run to any command to see the exact credit price first (0 credits, generates nothing).


Contents

What is Distribea

Most AI media tools hand you a raw model and a blank prompt box, in a browser tab, one image at a time. Distribea is the layer above that — and it lives where you build.

  • One connector, four media types. Image, video, music and voice through a single install. Most tools stop at images.
  • Three ways in, one engine. Run it from your shell (distribea …), drop it into your AI agent as an MCP server (Claude Code, Cursor, Windsurf, Replit, Copilot), or add the Claude Skill so the model reaches for it on its own.
  • On-brand by default. Lock your brand once and every asset matches it. The art director turns a one-line subject into a full prompt — you never write the "how".
  • Production-ready. Images come out as optimised WebP with proper SEO alt text, downloaded for you or patched straight into your page.

The heavy lifting — art direction, generation, billing — runs on the hosted Distribea engine. This package is only the local connector: it scans and patches your files, holds no API keys and no secrets, and bills generation in credits on your Distribea plan at the regular site price.

What you can make

MediaCommandWhat you get
🖼️ Imagedistribea image "<subject>"One on-brand image (png/jpg), any aspect ratio. Portraits, heroes, product shots, illustrations, review avatars.
🎬 Videodistribea video "<subject>"One short clip (mp4) — text-to-video, or image-to-video from a reference photo.
🎵 Musicdistribea music "<style>"One royalty-ready track (mp3) — intros, backing music, ad beds.
🎙️ Voicedistribea voice "<text>"One voice-over (mp3), text-to-speech in many languages.
📄 A whole pagedistribea page [path]Every <img> placeholder on a page (or the whole project) filled on-brand, in parallel, patched into your code.
✍️ Blog coverdistribea blog "<article>"A 16:9 cover image that illustrates the article's actual subject, in your style, + alt text.

When to use Distribea

Reach for Distribea whenever a project needs imagery or audio and you'd otherwise ship stock photos, gray placeholders, CSS-gradient-only sections, or a raw pasted photo. Typical jobs:

  • You're building a landing page and need real, on-brand images — not stock.
  • You need e-commerce product shots that stay consistent across the catalog.
  • You want a hero image, team portraits, or testimonial avatars with believable faces.
  • You're writing a blog post and need a cover image matched to the article and your style.
  • You need a short video clip (text-to-video or image-to-video) for a hero or ad.
  • You need background music for an intro, demo, or ad.
  • You need a voice-over for a demo, walkthrough, or video narration.
  • You have a site full of cheap or off-brand stock photos and want to rebrand them in place.
  • You're vibe coding with an AI agent and want it to fill in media on its own as it builds.

Why Distribea

Raw-model CLI / MCPGeneric image MCPDistribea
Image · video · music · voice⚠️ varies❌ image only✅ all four, one connector
Lives in your terminal & your agent⚠️ one of them⚠️ MCP only✅ CLI · MCP · Skill
Writes the art direction for you❌ you prompt each model⚠️ prompt by prompt✅ built-in art director
Locks your brand once❌⚠️ per prompt✅ applied everywhere
Same face across scenes❌❌✅ recurring characters
Fills a whole page in one call❌❌✅ distribea page
Production-ready (WebP + SEO alt)❌❌✅ automatic
No keys/secrets in the connector⚠️ holds your model keys⚠️ varies✅ browser sign-in, nothing to paste

In one line: other tools give you a model and a prompt box; Distribea gives you finished, on-brand assets — in the four media types you actually need — without leaving the place you build.

Install

A Distribea plan powers generation (free/trial accounts get a couple of gift images to try it). The connector itself is free and stores no secrets.

CLI (recommended)

bash
npm install -g distribea-mcpdistribea login          # opens your browser, click Authorize — no key to paste

distribea whoami confirms you're connected.

MCP server

Drop this into the .mcp.json at the root of your project:

json
{  "mcpServers": {    "distribea-mcp": {      "command": "npx",      "args": ["-y", "distribea-mcp@latest"],      "env": {        "DISTRIBEA_MCP_KEY": "dmcp_…"      }    }  }}

Grab your dmcp_… key at distribea.com/account/mcp. If you've already run distribea login, the server reuses that session and you can omit the env block entirely.

ToolWhere to add it
Claude Codeclaude mcp add distribea-mcp -e DISTRIBEA_MCP_KEY=dmcp_… -- npx -y distribea-mcp@latest
CursorSettings → MCP → Add, or paste the block into ~/.cursor/mcp.json
WindsurfPaste into ~/.codeium/windsurf/mcp_config.json
VS Code (Copilot)Paste into .vscode/mcp.json
ReplitAdd as a custom MCP server with the command above

Claude Skill

The package ships SKILL.md. Add it once and your agent reaches for Distribea on its own whenever a page needs imagery or audio.

bash
npx skills add Distribea/distribea-mcp

Quickstart

bash
npm install -g distribea-mcpdistribea logindistribea image "a quiet beach at sunrise, warm light" --landscape

The file lands in a tidy folder, ready to use. That's it — no model to pick, no prompt engineering: the style is inferred and the art director writes the prompt. From here, swap image for video, music, or voice, or run distribea page to dress a whole page at once.

Examples & recipes

Every command takes a plain-language subject. Add --dry-run to preview the price first (0 credits, generates nothing).

On-brand image

bash
distribea image "founder portrait, soft daylight, neutral background" --portrait

Standalone video (text-to-video)

bash
distribea video "slow waves rolling onto a tropical beach" --landscape

Background music

bash
distribea music "warm lo-fi beat, mellow, 90 bpm" --out ./assets

Voice-over (text-to-speech)

bash
distribea voice "Welcome to our store — let's get you started." --lang en

Reference photo → image or video (image-to-image / image-to-video)

Turn a real photo into a new on-brand image or animate it. Pass --ref once, or several times for multiple references — automatically capped to what the chosen model accepts.

bash
distribea image "the same product on a marble kitchen counter" --ref ./bottle.jpgdistribea video "gentle push-in on the cliff at golden hour" --ref ./cliff.jpg --portrait

Blog cover

Reads the article and illustrates its specific subject, matched to your locked style, as a 16:9 WebP + alt text.

bash
distribea blog "How we cut our build times in half — full article body here…"

Pick a specific model

Defaults are strong; only name a model when you want a particular one.

bash
distribea image "minimalist blue logo on white" --model nano-banana-pro --dry-run

Browse the live catalog (free)

bash
distribea models image      # or: video | music | voice

Dress a whole page

The page workflow is the superpower. Build your page with <img src="https://placehold.co/1200x600"> markers wherever an image goes, then one call fills every slot with imagery that matches your brand — one consistent look across the whole site, recurring characters with the same face in every scene, believable UGC avatars for review sections.

bash
distribea page ./index.html      # omit the path to dress the whole project

Every image lands production-ready: optimised WebP, proper SEO alt text, and patched straight into your code. No stock photos. No placeholders. No export step. The style is inferred from the page itself, so there's nothing to set up.

Already have a site with cheap or off-brand stock photos? Rebrand them in place — the first pass lists every replaceable image for free, then applies the new look without touching your code (originals kept as *.original).

Recurring faces & products

A real person (a founder, a teacher) or a product that must look identical in every shot is treated as a locked subject — the face/object stays 100% identical, only the staging changes.

bash
distribea image "the teacher explaining at a whiteboard" --ref ./founder.jpg

Never paste a raw photo into the page: leave a placeholder and let the engine restage that exact face in your site's world. Review and testimonial avatars come out as casual smartphone selfies (UGC look); the same reviewer keeps the same face across the site.

Models

Generation runs on best-in-class providers, with a strong on-brand default per media type — so you rarely need to name one:

TypeDefaultA few of the providers available
ImageNano Banana Pro (2K)Nano Banana Pro · Seedream · Z-Image · Grok Image
VideoGrok Imagine (6s, 720p)Grok · Seedance · Veo · Sora
MusicMiniMax MusicMiniMax · CassetteAI
VoiceElevenLabs Multilingual v2ElevenLabs · MiniMax

The catalog is live — browse current models, what each is best at, their options and indicative prices with distribea models image|video|music|voice (free), or list_models / list_video_models / list_music_models / list_voice_models over MCP.

Command reference

CommandPurpose
distribea login [key]Connect — opens your browser; or pass a dmcp_… key directly
distribea whoami · logoutCheck / forget the saved session
distribea image "<subject>"One downloadable image (png/jpg)
distribea video "<subject>"One downloadable video (mp4)
distribea music "<style>"One downloadable track (mp3)
distribea voice "<text>"One downloadable voice-over (mp3)
distribea page [path]Fill a page (or the whole project) — placeholders → on-brand images
distribea blog "<article>"Article cover (16:9 + alt text)
distribea models [type]Browse the live catalog: image · video · music · voice
distribea call <tool> --k vCall any advanced MCP tool directly (see MCP tools)

Run distribea help for the full reference.

Flags

FlagPurpose
--model <id>Pick a specific model (see distribea models)
--ref <file>Reference image — image-to-image / image-to-video (repeat for several)
--format <r>16:9 (landscape) · 9:16 (portrait) · 1:1 (square)
--portrait · --landscape · --squareOrientation shortcuts
--lang <code>Voice language (e.g. en, fr)
--out <dir>Where to download the file
--jpgImage as jpg instead of png
--dry-runQuote only (0 credits) — shows the price, generates nothing

MCP tools

When used as an MCP server, the same engine exposes these tools to your agent. The agent picks the right one for the job; you just describe what you need.

ToolWhat it does
make_images ⭐Fill every placeholder/stock slot of a page (or the whole project) with on-brand images, in parallel; or rebrand existing images in place (rebrand: true)
bring_aliveMake an existing site feel alive: add images where sections have none and replace stock/old pictures on-brand (free proposal first, reversible)
generate_imageOne on-brand image (auto-switches to UGC selfie mode for review avatars)
generate_with_modelStandalone image with a chosen model + settings, downloaded as a native png/jpg
generate_videoStandalone video with a chosen model, downloaded as a native mp4
generate_musicStandalone music track, downloaded as mp3
generate_voiceStandalone voice-over (text-to-speech), downloaded as mp3
list_models · list_video_models · list_music_models · list_voice_modelsBrowse the catalog (ids, strengths, options, indicative prices) before generating (free)
edit_imageRetouch, redo, remove background, upscale ×4, extend
blog_coverGenerate an on-brand cover image for a blog article
site_styleLock, refine or anchor the site's visual identity
brand_packLogo, favicon pack and og:image in one call
create_referenceLock a recurring character (real photo supported) or product
finish_imagesFix missing ALT texts and convert heavy images to WebP (free)
pack_statusCurrent style, characters, credits
list_projects · link_project · forget_projectSave, reconnect or wipe a project's brand memory (free)

How it works

  1. You give a subject — a CLI command, or your agent calls a tool while it builds.
  2. The art director expands it into a full, on-brand prompt using your locked style — so you never hand-write the "how".
  3. The best-fit model generates the asset (or the one you named with --model).
  4. You get a finished file — downloaded to a tidy folder, or patched into your page as optimised WebP with SEO alt text, originals kept safe.

Every batch is quoted in credits before it runs; --dry-run (CLI) or dry_run (MCP) previews the price and generates nothing. The connector runs locally and holds no secrets; your brand memory, generated assets, and billing live on the hosted Distribea engine.

Billing & cost

  • Generation is billed in credits on your Distribea plan; every batch is quoted in credits before it starts — use --dry-run (CLI) or dry_run (MCP) to preview the price first.
  • Video, music and voice need a paid plan and take longer to render. If a call says it's still rendering, run it again with the same settings to pick it up — you are never charged twice.
  • One page call dresses a page; the job is then done. Distribea won't regenerate or "improve" assets on its own — retouch only when you ask.

Privacy & security

This local connector contains no API keys and no secrets. To generate an asset it sends to the hosted Distribea engine only what is needed for the request:

  • What is sent: your brief (the subject you ask for), a short text excerpt of the page being worked on (to infer the site's style and context), a hashed project identifier, and any file you explicitly ask to edit. Your other project files stay on your machine — they are never read or uploaded.
  • What is stored, tied to your account: your locked visual style, recurring characters/products/avatars, the assets generated for you (hosted on Distribea's CDN), and your credit/billing usage.
  • Third parties: assets are produced through Distribea's hosted AI providers solely to fulfil your request. Your data is never sold.
  • Authentication: calls are authenticated with your personal key (dmcp_…), issued and revocable at distribea.com/account/mcp.
  • Retention & contact: data is kept for the life of your account; full policy and contact at distribea.com/legal/privacy-policy.

FAQ

What is Distribea MCP? An on-brand AI media toolkit that generates images, video, music and voice from plain-language prompts, available as a CLI, an MCP server, and a Claude Skill. It ships production-ready assets (WebP + SEO alt text) at the regular site price.

Which AI coding tools does it work with? Claude Code, Cursor, Windsurf, Replit and GitHub Copilot — anything that speaks MCP. It also runs as a plain CLI in any terminal.

Do I need an API key? No. distribea login signs you in through the browser — nothing to paste. For an MCP/hosted setup you use a personal dmcp_… key, issued and revocable in your account.

Is it free? The connector is free and stores no secrets. Generation is billed in credits on your Distribea plan; free and trial accounts get a couple of gift images to try it.

Can it generate video, music and voice — or only images? All four: image, video, music and voice, through one connector.

How is it different from a raw image model or a generic image MCP? A built-in art director writes the prompt from your one-line subject, your brand is locked once and applied everywhere, faces and products stay identical across shots, and a single call can fill an entire page — patched into your code as WebP with SEO alt text.

Can it use my own photos? Yes. Pass --ref for image-to-image or image-to-video, and lock a person or product so the same face/object appears in every scene while the staging changes.

Will it modify my code or files? Only files you point it at. distribea page fills placeholders and patches them into your markup; when rebranding existing images it keeps the originals as *.original. Your other files are never read or uploaded.

What formats and aspect ratios does it produce? Downloaded assets are native png/jpg (image), mp4 (video), mp3 (music/voice); images placed in a page come out as optimised WebP. Orientations: 16:9, 9:16, 1:1.

How do I see the price before generating? Add --dry-run (CLI) or dry_run (MCP). It quotes the exact credit cost and generates nothing (0 credits).

A call says it's still rendering — will I be charged twice? No. Video, music and voice take longer; re-run the exact same command to pick the result up. You are never charged twice.

Which models does it use? Strong on-brand defaults per media type (e.g. Nano Banana Pro for images, Grok Imagine for video, MiniMax for music, ElevenLabs for voice), with a live catalog you can browse and override with --model.

Troubleshooting

No Distribea key / Not authenticated — run distribea login (or set DISTRIBEA_MCP_KEY for an MCP/hosted setup).

Unknown model "<name>" — run distribea models image|video|music|voice for the current catalog.

A call says it's still rendering — video/music/voice take longer; re-run the exact same command to pick the result up. You are never charged twice.

Updating

bash
npm install -g distribea-mcp@latest

The MCP server pulls @latest automatically through npx, so agent setups stay current on their own.

Support

Bugs and feature requests: github.com/Distribea/distribea-mcp/issues. Key and docs: distribea.com/account/mcp.

License

MIT

© Distribea · distribea.com

来源:README.md,提交 a8712ea

工具

0
该服务器需要登录,因此 SourceWeft 无法在没有账号的情况下读取其工具,也未收录。已连接账号所能看到的内容可能因账号及其权限而异。

版本历史

1
  1. v2.1.0最新Sep 16, 2026

评分

已满足适用检查项的 85%

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 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 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