Orcadub Mcp Server

io.github.Continuum-AI-Corp由 GitHub 账号 Continuum-AI-Corp 发布v1.4.0更新于 Oct 24, 2026

AI video dubbing via OrcaRouter (model orca/dub): upload a video or URL, poll, download the MP4.

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

OrcaDub 官方 MCP 服务器,让助手通过 OrcaRouter 的 orca/dub 模型把视频配音成其他语言,并下载生成的 MP4。

可以做什么
  • 检查网关与模型路由是否正常,不创建任务也不计费
  • 上传本地视频文件(超过 64 MiB 自动分块,最大 8 GiB)并获得 file_id
  • 用 file_id 或 URL 提交配音任务,指定源语言和目标语言,可选术语表、翻译风格、口型同步、水印和分辨率等参数
  • 轮询任务状态与进度,直到完成或失败,并获取交付地址
  • 把完成的 MP4 保存到本地路径,且不会覆盖已有文件
适用场景
适合希望由助手把视频中的语音翻译成受支持语言并保留原声的场景,可在 Claude Code、Claude Desktop、Codex CLI、Cursor、Windsurf 等支持 stdio 的 MCP 客户端中使用。
运行要求
以本地 stdio 进程运行,通常通过 npx @orcadub/cli、Docker 镜像、预编译二进制或源码构建启动;使用 npx 方式需要 Node.js。需要 OrcaRouter API 密钥,通过环境变量 ORCADUB_API_KEY 提供,并需能访问固定的 OrcaRouter 网关。上传和下载工具会读写本地路径,容器方式需挂载相应目录。
安装前请注意
每次工具调用都用 ORCADUB_API_KEY 认证,网关据此确定工作区和计费。配音任务按源视频分钟数计费,dub_create 会产生费用,需要账户有余额。上传的文件和源链接会发送到 OrcaRouter 网关处理。dub_download 会写入本地文件,返回的 content_url 是不过期的 Bearer 认证链接。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Orcadub Mcp Server,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

[OrcaDub]

Translate any video. Keep the voice.

The official MCP server for OrcaDub — AI video dubbing, driven from your agent.

[CI] [Release] [npm] [License: MIT]

Website · API Docs · Get an API key · Config examples · Releases


Give any MCP-capable agent — Claude Code, Claude Desktop, Codex CLI, Cursor, Windsurf — the ability to dub a video into another language: upload a file or pass a URL, submit a job to the orca/dub model through the OrcaRouter gateway, poll progress, and download the finished MP4.

CLI usage

Every operation is a one-shot subcommand (no resident server needed):

bash
export ORCADUB_API_KEY=sk-orca-...   # from https://www.orcarouter.ai/console
npx -y @orcadub/cli healthnpx -y @orcadub/cli upload --path ./clip.mp4npx -y @orcadub/cli create --source-lang en --target-lang ja \  --url https://youtu.be/... --opt preserve_bgm=truenpx -y @orcadub/cli get --video-id <id>npx -y @orcadub/cli download --video-id <id> --dest ./out.mp4

Optional create parameters use repeatable --opt key=val (e.g. --opt watermark=false --opt resolution=1080p --opt glossary.OrcaDub=OrcaDub). Results print as JSON on stdout; errors go to stderr with a non-zero exit.

With no subcommand the same binary runs as an MCP stdio server (npx -y @orcadub/cli).

Install the agent Skill

The dub-video Skill teaches an agent when to dub, which billed parameters it must confirm, and how to run the upload → create → poll → download workflow. Install it interactively:

bash
npx -y @orcadub/cli skill install

The guided installer opens with a balanced 8-row ORCADUB wordmark, then lets you:

  • Choose the Simplified Chinese or English interface; the system locale selects the initial default.
  • Choose Project or Global installation with the arrow keys.
  • Select one or more platforms from a checkbox list. Detected platforms appear first and are already checked.
  • Press Space to toggle, / to filter, a to select all, n to clear all, and Enter to confirm.

It supports 36 platforms: the 33-platform Comet-compatible catalog plus Hermes, OpenClaw, and Command Code. Use --lang zh or --lang en to skip the language screen. For unattended or agent-driven setup, select targets explicitly:

bash
# Current projectnpx -y @orcadub/cli skill install \  --platform claude --platform codex --scope project --lang en --yes
# User-wide Codex installationnpx -y @orcadub/cli skill install \  --platform codex --scope global --lang zh --yes

Use --yes to accept detected platforms (or all platforms when none are detected) without opening the installer. Use --json for decoration-free structured output. Set NO_COLOR (or use TERM=dumb) for a color-free installer. Existing identical content is left unchanged; an existing different Skill is preserved unless --force is provided. Skill installation needs network access to the canonical orcadub-plugin repository, but it does not require ORCADUB_API_KEY and does not contact OrcaRouter.

Quick start

bash
claude mcp add orcadub -e ORCADUB_API_KEY=sk-orca-... -- npx -y @orcadub/cli

Then just ask your agent: "Dub this into Chinese: https://www.youtube.com/watch?v=…"

Prerequisites

An OrcaRouter API key is required — every tool call is authenticated:

  1. Register / log in at the OrcaRouter console.
  2. Create an API key on the token management page (it looks like sk-orca-...).
  3. Provide it via the ORCADUB_API_KEY environment variable (see below).

Dubbing jobs are billed per minute of source video. Without a key the server still starts, but every tool call returns a sign-up redirect to the console.

Installation

Claude Code

bash
claude mcp add orcadub -e ORCADUB_API_KEY=sk-orca-... -- npx -y @orcadub/cli

Claude Desktop

Add to claude_desktop_config.json:

json
{  "mcpServers": {    "orcadub": {      "command": "npx",      "args": ["-y", "@orcadub/cli"],      "env": { "ORCADUB_API_KEY": "sk-orca-..." }    }  }}

Codex CLI

bash
codex mcp add orcadub --env ORCADUB_API_KEY=sk-orca-... -- npx -y @orcadub/cli

Cursor / Windsurf / other MCP hosts

Any host that launches stdio MCP servers works with the same shape: command npx, args ["-y", "@orcadub/cli"], env ORCADUB_API_KEY. Ready-to-copy configuration files for Claude Code, Codex, Cursor and Windsurf live in examples/.

Docker

Images are published to GHCR for linux/amd64 and linux/arm64:

bash
docker pull ghcr.io/continuum-ai-corp/orcadub-mcp-server:latestclaude mcp add orcadub -e ORCADUB_API_KEY=sk-orca-... -- \  docker run --rm -i -e ORCADUB_API_KEY -v "$PWD:$PWD" -w "$PWD" \  ghcr.io/continuum-ai-corp/orcadub-mcp-server:latest

Or as host JSON config:

json
{  "mcpServers": {    "orcadub": {      "command": "docker",      "args": ["run", "--rm", "-i", "-e", "ORCADUB_API_KEY",               "-v", "/path/to/videos:/work", "-w", "/work",               "ghcr.io/continuum-ai-corp/orcadub-mcp-server:latest"],      "env": { "ORCADUB_API_KEY": "sk-orca-..." }    }  }}

Note: dub_upload and dub_download read/write local paths, so mount the directory you upload from / download to (the -v flag) — paths you pass to the tools must be valid inside the container.

Prebuilt binaries

Download the binary for your platform from the releases page (verify with checksums.txt), then register it directly:

bash
# example: Linux amd64 via gh CLIgh release download v0.1.0 -R Continuum-AI-Corp/orcadub-mcp-server \  -p 'orcadub-mcp-server_*_linux_amd64' -O ~/bin/orcadub-mcp-server && chmod +x ~/bin/orcadub-mcp-serverclaude mcp add orcadub -e ORCADUB_API_KEY=sk-orca-... -- ~/bin/orcadub-mcp-server

From source

bash
git clone https://github.com/Continuum-AI-Corp/orcadub-mcp-server && cd orcadub-mcp-server && make build# binary lands in bin/orcadub-mcp-server

Configuration

Environment variableRequiredDescription
ORCADUB_API_KEYyesOrcaRouter sk-orca-... key, sent as Authorization: Bearer on every request. The gateway resolves your workspace and billing from it.

The gateway address (https://api.orcarouter.ai) is fixed in the binary — there is nothing else to configure.

Tools

ToolDescription
dub_healthEnd-to-end probe of gateway → orca/dub routing. No job is created, nothing is billed.
dub_uploadUpload a local video file; returns the file_id for dub_create. Files above 64 MiB are chunked automatically (up to 8 GiB).
dub_createSubmit a dubbing job (billed). Source is exactly one of file_id or url (YouTube and direct links are fetched server-side). Required: source_lang, target_lang (never auto), and video_name when using file_id. Optional knobs cover content profiles, glossaries, translation style, background-music preservation, lipsync, watermarking, resolution/ratio and more.
dub_getPoll a job: queued → in_progress → completed | failed with integer progress. On completion the response includes content_url (Bearer-authenticated delivery address, never expires) and job_id.
dub_downloadSave a completed job's MP4 to a local path (refuses to overwrite existing files).

Supported languages: en zh ja ko fr de es pt ru ar it hi tr th vi id bn pl nl uk fil el cs sv da no fi sk.

Example

A typical agent conversation:

User: Dub this into Chinese: https://www.youtube.com/watch?v=xxxx

Agent: (dub_create with source_lang=en, target_lang=zh, url=…) → job queued → (polls dub_get every ~30 s) → completed → (asks: download locally? default: current directory) → (dub_download) → "Saved to ./My-Video-zh.mp4 — re-download any time: curl -H "Authorization: Bearer sk-orca-..." <content_url> -o out.mp4"

Direct tool usage from an agent prompt works too — for example dub_create {"source_lang":"en","target_lang":"zh","url":"https://…","glossary":{"OrcaRT":"OrcaDub Realtime"}}.

Troubleshooting

SymptomCause / fix
Tool calls return not authorized … www.orcarouter.ai/consoleORCADUB_API_KEY is unset — register, export the key, restart the session.
401Invalid or expired key — re-issue on the OrcaRouter console.
402 insufficient_creditTop up your OrcaRouter balance (per-minute billing).
429 free_quota_exceededFree-tier limit reached.
task_not_exist on dub_getThe id wasn't created through this gateway — use the id exactly as returned by dub_create.

Development

bash
go build ./...go test ./... -count=1

Layout: cmd/ (entrypoint) · internal/ (HTTP client, Skill installer, tool layer, wire types) · npm/ (npx launcher that downloads the platform binary from GitHub Releases) · server.json (MCP Registry manifest).

Releasing

  1. Tag: git tag v0.1.0 && git push origin v0.1.0.
  2. GitHub Actions runs goreleaser: linux/darwin/windows × amd64/arm64 binaries + archives + checksums land on the Release.
  3. The npm-publish job stamps the tag version into npm/package.json and publishes @orcadub/cli to npm (requires the NPM_TOKEN repo secret).
  4. The registry job stamps the same tag version into server.json and publishes it to the MCP Registry.

License

MIT

来源:README.md,提交 14de090

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. v1.4.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

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