each::labs

ai.eachlabs由 eachlabs.ai 发布官方发布者v0.3.0更新于 Oct 29, 2026

AI media built for production. AI orchestration for image, video and audio, in any AI app.

已验证Streamable HTTP可网页运行效率与工作流媒体与设计

概览

AI 生成的概览

each::labs 远程连接器，让助手生成和编辑图像、视频、音乐与语音，并运行所属组织的媒体工作流。

可以做什么
  • 根据提示生成图像、视频、音乐和语音，可一次出多个变体供挑选
  • 编辑媒体：去背景、照片风格化、让静态图动起来、转写音频
  • 剪辑、裁剪、缩放和转换片段，提取音频或转成 GIF
  • 列出、查看并运行所属组织的 each::labs 工作流，展示其输出
  • 在对话中显示交互卡片，用于上传文件、调整模型设置和预览结果
适用场景
适合在 Claude、Codex 或 Grok Build 的对话中直接完成偏生产用途的媒体生成与编辑，或团队已有 each::labs 工作流、希望不离开聊天就能触发时使用。
运行要求
通过 streamable HTTP 连接 mcp.eachlabs.ai 远程端点，需用 each::labs 账号完成 OAuth 登录；无需 API 密钥或本地运行时。团队场景下由 Owner 添加连接器，每位成员用应计费组织内的账号登录。
安装前请注意
每次模型和工作流运行都计入登录时选择的组织，用个人账号登录会改为向个人计费。提示词和输入文件链接会发送到 each::labs，卡片上传的文件由浏览器直接存入 each::labs 存储，任何拿到链接的人都能打开；文件保留 180 天或按组织设定的期限。变体选定后不可更改。

安装

在 SourceWeft 中

  1. 打开 控制台中的 each::labs，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/ai.eachlabs%2Fmcp/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": {
    "each-labs": {
      "type": "http",
      "url": "https://mcp.eachlabs.ai/mcp"
    }
  }
}

README

each::labs plugin for Claude, Codex and Grok Build

Generate and edit images, video, music and voice with each::labs models, and run your organization's each::labs workflows, by asking in a conversation. The plugin also cuts, crops, resizes and converts media, and shows small interactive cards inside the chat where you upload a file, adjust a model's settings or pick one of several results.

It works in Claude Chat (claude.ai), Claude Desktop, Cowork and Claude Code, and in the Codex app, Codex CLI and Grok Build. You sign in with your each::labs account; there is no API key to paste.

Install

The plugin is installed from its own marketplace, eachlabs/plugin. The same repository serves Claude, Codex and Grok Build.

WhereHow
Claude Chat, Claude DesktopCustomize, Plugins, Add marketplace, enter eachlabs/plugin, then add the eachlabs plugin
Claude Codeclaude plugin marketplace add eachlabs/plugin, then claude plugin install eachlabs@eachlabs
Codex CLIcodex plugin marketplace add eachlabs/plugin, then codex plugin add eachlabs@eachlabs, then codex mcp login eachlabs
Codex appAdd the marketplace with the CLI command above, then install the eachlabs plugin from Plugins
Grok Buildgrok plugin install eachlabs/plugin --trust, then sign in when Grok opens the each::labs page in your browser

Team and Enterprise. An Owner can add the connector https://mcp.eachlabs.ai/mcp for the whole organization. Each member adds the plugin marketplace themselves, because organization-wide plugin sync on claude.ai accepts only the organization's own private repositories. Each member then signs in with their own each::labs account. That account must belong to the each::labs organization that should pay for the runs; a member who signs in with a personal account is billed there instead.

Updating. In Claude Chat and Claude Desktop a plugin installed from the marketplace updates automatically, with no restart. In Claude Code, run claude plugin update eachlabs@eachlabs, then /reload-plugins. In Codex, run codex plugin marketplace upgrade. In Grok Build, run grok plugin update eachlabs.

Sign in

After installing, open the plugin's Connectors tab and choose Connect. A browser window opens on each::labs: sign in and pick the organization to use. Every model run and workflow run is billed to that organization, the same way as a run from the each::labs dashboard or API.

In Grok Build, the browser opens the first time the plugin is used. To sign in again, open /mcps and press i on eachlabs.

What to ask

Ask in your own words. Claude chooses the model, reads its inputs and runs it. Some examples:

TaskExample
Images"Generate a product photo of a white sneaker on a marble table, soft studio light."
Images"Give me 4 variants of a fantasy trading card with a fire dragon, I want to pick one."
Image editing"Remove the background from this photo." "Turn this portrait into a watercolor painting."
Video"Turn this image into a 5-second clip where the camera slowly pushes in."
Video"Make a 10-second video of waves hitting a lighthouse at sunset."
Music"Compose a 30-second upbeat lo-fi beat for an intro."
Voice"Read this paragraph aloud in a calm male voice." "Transcribe this interview."
Cutting and cropping"Cut the first 10 seconds of this video." "Crop this video to 9:16 for a story."
Converting"Convert this clip to a GIF." "Extract the audio from this video as mp3."
Workflows"Which workflows do we have?" "Run our product clip workflow with this photo."

When a task has several steps, such as restyling a photo and then animating it, Claude first looks for one of your organization's workflows that does the whole job, and runs it if one fits.

Cards in the conversation

In the chat apps, Claude Chat (claude.ai), Claude Desktop and the Codex app, some answers come with a small card instead of plain text. Terminals such as Claude Code, Codex CLI and Grok Build show the same results as text.

You seeWhen
A file picker with "Choose a photo" and a progress barA model needs a file you have not linked. One card takes a whole set of files, for example all the screenshots for a slideshow. The files go from your browser straight to each::labs storage
A form with the model's settings and a Run buttonYou ask to adjust settings, or a run has options worth choosing first
The result: the image, a video or audio player, and download buttonsEvery run that produces media. A file in several formats has one button per format. A workflow's outputs show together as a gallery
Several results to choose from, and a "Use this variant" buttonYou asked for several variants. Click between them as often as you like, then press "Use this variant"; Claude continues with that one, and the choice is final
A Compare button under an edited image or videoThe result is an edit of a file you gave. Drag the handle to compare before and after
A timeline with two handles, or a rectangle on the first frameYou want to cut a clip or crop a frame and have not given exact numbers
A workflow's inputs as a form, with a file picker for each photo, video or audio inputA workflow needs inputs you have not given

After you press a card's button, Claude continues on its own. You do not need to type anything. A card you already used stays locked on your choice when you come back to the conversation later.

After the first run of a model in a conversation, Claude ends its reply with a short Adjust generation settings line naming a few of that model's settings. Ask to change any of them and the settings form appears.

Limits

  • each::sense skills are not included. The public eachlabs/skills collection also has skills that run through each::sense, the each::labs agent. Access to each::sense is restricted to specific organizations, so this plugin does not include those skills. Most of their tasks, such as background removal, upscaling, trimming or watermarking, work through the models and video tools above.
  • Files live on each::labs storage. Uploads and results are stored by each::labs and shown from there. Anyone who has a file's link can open it; the links are long random addresses, not listed anywhere. Files are kept for 180 days, or for your organization's own retention period if it set one, and can be deleted earlier from the storage page in the each::labs dashboard.
  • Uploads. One upload card takes up to 50 files, or fewer when the chosen model accepts fewer, and up to 100 MB per file. Files upload four at a time. A card refuses too many files or a file that is too large before anything uploads. The same limits apply to the file fields on a workflow's form.
  • Long jobs. A video can take a few minutes. Claude waits for it; if a run takes longer than the wait, Claude checks again until it finishes.
  • Variants. Up to 8 results of one prompt in a single request, unless the model itself can make more in one run.
  • Workflows. Claude can list, read and run your organization's workflows and show their results. Building or editing a workflow is done in the each::labs dashboard.
  • Phones. Cards work in the Claude app; the file picker opens your photo library, not the camera.
  • Trimming FLAC audio returns an mp4 file, because the trim copies the audio without converting it. mp3, wav and m4a keep their format.

Troubleshooting

ProblemWhat to do
A new card or an update does not showStart a new conversation. If it still does not show, reload claude.ai or restart Claude Desktop
A card says it takes fewer files than you haveTell Claude how many files you have, or ask for another card for the rest
Tools fail with a sign-in or permission errorOpen the plugin's Connectors tab, disconnect, and connect again
A card does not appear, only textCheck you are in Claude Chat or Claude Desktop; Claude Code shows text only. Ask Claude to show the result again
A card says a file is no longer availableAsk Claude to show the file again; it fetches a fresh copy of the result
The wrong organization is billedDisconnect the connector and sign in again, picking the right organization

Data

The plugin connects to one address, https://mcp.eachlabs.ai/mcp, and needs only an each::labs sign-in: no API key, and no scripts or hooks run on your machine.

Your prompts and input file links go to the each::labs MCP server, which runs them on the each::labs API under the organization you picked at sign-in. A file you add on a card goes from your browser to each::labs storage and does not pass through Claude. The plugin stores nothing itself.

Contributing

How the repository is laid out, how the skills are generated and how a version is released is in CONTRIBUTING.md.

来源：README.md，提交 1cf56db

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. v0.3.0最新Oct 29, 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