Loom MCP 4fc5636a

MCP音视频

提供方 karbassi

音视频 stdio · 本地进程 会话登录 社区 / 第三方 Claude DesktopClaudeCursorVS Code
查看接入方式

Loom MCP服务器提供59种工具,用于管理Loom内部GraphQL API的视频内容,包括列表、搜索、元数据获取、转录、注释、评论、任务和标签管理等功能。

一句话选型结论

按资料字段整理

Loom MCP 4fc5636a 属于音视频类 MCP 服务器,stdio · 本地进程,会话登录,可直接接入 Claude Desktop、Claude、Cursor 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类音视频 传输方式stdio · 本地进程 鉴权会话登录 部署方式本地部署
GitHub 星标 4 仓库热度,无公开仓库时显示未标注
工具数 59 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

音视频
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量59暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

视频管理Python本地部署GraphQLAPI内容提取协作工具自动化工具

安装与配置

stdio · 本地进程 会话登录
客户端在本机拉起子进程,数据不出本机
uv run loom-mcp

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Python,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后客户端会列出约 59 个工具,随便调一个只读工具确认能返回结果。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
转写与字幕

把音视频转成文字和字幕。

音视频处理

裁剪、转码、拼接素材。

内容分发

把成品发到对应平台。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 资料标注 59 个工具,覆盖面较广,能满足多种调用。
  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 使用前要准备好会话登录对应的凭据,并按最小权限配置。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在音视频场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Loom MCP 4fc5636a当前 59 stdio · 本地进程 本地部署 有 2026-05-17
Ableton MCP B7b69a6a 16 stdio · 本地进程 本地部署 有 2026-05-17
Mini Max MCP 8 SSE · 服务端推送 远程部署 有 2026-05-18
LocalAI 未标注 stdio · 本地进程 本地部署 有 2026-04-11
Youtube E0eb3510 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Voicemode MCP 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

音视频
字段来自项目仓库与公开索引,中文说明由本站标注
提供方karbassi上游标注的维护方 karbassi
分类音视频本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式会话登录远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · Python本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

9.3 KB
来自项目仓库与上游目录的原文,未做删改

Loom MCP Server

MCP server exposing 59 tools for Loom's internal GraphQL API. Works with Claude, Cursor, or any MCP-compatible client.

Features

  • List, search, and get detailed metadata for Loom videos
  • Read transcripts, captions, AI summaries, chapters, and key takeaways
  • Save any fetched content to disk on demand via save_dir parameter
  • Manage comments, tasks, reactions, and tags
  • Organize with folders, spaces, and watch lists
  • Update video settings, share to spaces, and more

Prerequisites

  • Python 3.11+
  • uv package manager

Setup

Authentication

> [!IMPORTANT]
> This server uses Loom's internal GraphQL API via a browser session cookie. There is no official API key — you must grab the cookie from your browser.

  1. Open loom.com in your browser
  2. Open DevTools (F12) → Application → Cookies → https://www.loom.com
  3. Copy the value of the connect.sid cookie (starts with s%3A...)
  4. Paste it as LOOM_COOKIE in your MCP client config below, prefixed with connect.sid=

The cookie lasts about 30 days. See Troubleshooting if you get auth errors.

Installation

Pick your MCP client below for the appropriate config. Each uses uvx to install and run directly from GitHub — no clone needed.

Claude Desktop

Add to your claude_desktop_config.json:```json
{
"mcpServers": {
"loom": {
"type": "stdio",
"command": "uvx",
"args": ["--from", "git+https://github.com/karbassi/loom-mcp.git", "loom-mcp"],
"env": {
"LOOM_COOKIE": "connect.sid=s%3A..."
}
}
}
}


Claude Code```sh
claude mcp add loom -- uvx --from git+https://github.com/karbassi/loom-mcp.git loom-mcp

Then set the env var in your shell or .env:```sh
export LOOM_COOKIE="connect.sid=s%3A..."


Cursor

Add to your Cursor MCP settings (`.cursor/mcp.json`):```json
{
  "mcpServers": {
    "loom": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "git+https://github.com/karbassi/loom-mcp.git", "loom-mcp"],
      "env": {
        "LOOM_COOKIE": "connect.sid=s%3A..."
      }
    }
  }
}

VS Code

Add to your VS Code settings (.vscode/mcp.json):```json
{
"mcpServers": {
"loom": {
"type": "stdio",
"command": "uvx",
"args": ["--from", "git+https://github.com/karbassi/loom-mcp.git", "loom-mcp"],
"env": {
"LOOM_COOKIE": "connect.sid=s%3A..."
}
}
}
}


Local clone```sh
git clone git@github.com:karbassi/loom-mcp.git
cd loom-mcp
uv sync
uv run loom-mcp

Or reference it from any MCP client:```json
{
"mcpServers": {
"loom": {
"type": "stdio",
"command": "uv",
"args": ["run", "--directory", "/path/to/loom-mcp", "loom-mcp"],
"env": {
"LOOM_COOKIE": "connect.sid=s%3A..."
}
}
}
}


### Shared HTTP server

By default every client starts its own stdio process. To run one long-lived server that many clients share:```bash
LOOM_COOKIE="connect.sid=s%3A..." LOOM_MCP_TRANSPORT=http LOOM_MCP_PORT=8505 loom-mcp
``````json
{
  "mcpServers": {
    "loom": { "type": "http", "url": "http://127.0.0.1:8505/mcp" }
  }
}

The endpoint has no authentication: anything that can connect to it can use every tool with your Loom session, including deleting videos. So:

  • The server only binds to loopback (LOOM_MCP_HOST must be 127.0.0.1, localhost, or ::1). Any local process can still connect, so only use HTTP mode on a machine where every local user and process is trusted with your Loom account.
  • It always turns on FastMCP's Host/Origin protection, which FastMCP leaves off by default. That check only stops DNS-rebinding attacks from web pages. It does not authenticate anyone.
  • To reach it from another machine, put it behind a reverse proxy that controls who can connect: a bearer token, mTLS, an IP allowlist on a private network, or similar. Then add the proxy's hostname to FASTMCP_HTTP_ALLOWED_HOSTS, e.g. FASTMCP_HTTP_ALLOWED_HOSTS='["mcp.example.com"]'. If a browser-based client will call it, also add that client's origin to FASTMCP_HTTP_ALLOWED_ORIGINS, e.g. '["https://mcp.example.com"]'. Never expose the port or the proxy publicly without access control.

File-writing tools (save_dir on the read tools, download_media) write on the server's machine, not the client's, and need absolute paths. See Saving to disk.

Tools

Read (31 tools)

Tools marked with Save accept an optional save_dir parameter — see Saving to disk.

Tool Description Save
list_videos List your videos, sorted by most recent
search_videos AI-powered semantic search
get_video Video metadata (name, duration, owner, views) metadata.json
get_transcript Full transcript with timestamps and speakers transcript.txt
get_captions WebVTT captions with start+end timestamps per cue captions.vtt
get_summary AI-generated summary summary.txt
get_chapters AI-generated chapters chapters.txt
get_description AI-generated detailed description with timestamped sections description.txt
get_key_takeaways AI-generated key takeaways takeaways.txt
get_comments Comments and replies comments.txt
get_tasks AI-generated action items tasks.txt
get_reactions Emoji reactions reactions.txt
get_tags Video tags tags.txt
get_backlinks External references (where the video is shared/embedded) backlinks.txt
get_meeting_notes Confluence meeting notes URL
get_confluence_pages Linked Confluence pages
get_download_url Signed MP4 download URL
download_media Download audio or video to a local file (works when MP4 export is disabled; optional start/end trim; needs ffmpeg) out_path
get_video_details All-in-one: metadata + transcript + chapters + summary + comments + tasks details.md + all above
list_folders List your folders
list_spaces List your workspaces
get_space Space details (name, privacy, primary)
search_folders Search folders by name
get_folder Folder details
get_last_watch_time Last timestamp where you stopped watching
get_watch_later_count Number of videos in your Watch Later list
get_total_videos_count Total videos created by a user
get_frequent_reactions Your most-used emoji reaction types
get_comment_reactions Emoji reactions on a specific comment
get_user User profile by ID (name, email, company, avatar)
search_workspace_tags Search tags in your workspace

Write (29 tools)

> [!NOTE]
> Write tools modify your Loom data. Destructive actions (delete, archive, move) cannot always be undone.

Tool Description
update_video_name Rename a video
update_video_description Update video description
update_video_settings Update video settings (downloads, comments, etc.)
create_comment Post a comment (with optional timestamp)
edit_comment Edit an existing comment
delete_comment Delete a comment
create_task Create an action item on a video
update_task Update the content of an action item
delete_task Delete an action item
approve_task Mark a task as approved
respond_to_task Respond to a task
add_reaction Add an emoji reaction at a timestamp
delete_reaction Delete a reaction
add_comment_reaction React to a comment with an emoji
toggle_following Follow/unfollow a video
toggle_following_tag Follow/unfollow a workspace tag
archive_videos Archive or unarchive videos
duplicate_video Duplicate a video
delete_video Permanently delete a video
recover_video Recover a deleted video from trash
pin_video Pin or unpin a video in your library
add_to_watch_later Add to Watch Later list
remove_from_watch_later Remove from Watch Later list
create_folder Create a new folder
rename_folder Rename a folder
delete_folders Delete folders
move_videos Move videos to a different folder
move_folders Move folders into a different parent folder
share_videos_to_spaces Share videos to one or more spaces

Saving to disk

Most per-video read tools accept an optional save_dir parameter. When provided, the tool saves its output to {save_dir}/{video_id}/ and returns the file path alongside the content. When omitted, nothing is saved.

save_dir (and download_media's out_path) must be an absolute path; ~ is expanded. Relative paths are rejected because they would resolve against the server process's working directory, which means nothing for a shared or launchd-managed server.

Just ask naturally — "get the details from my last meeting and save them to ~/Downloads/loom" — and the LLM will pass save_dir="~/Downloads/loom" to the tool.

get_video_details saves each piece individually (metadata.json, transcript.txt, summary.txt, etc.) plus a combined details.md.

Troubleshooting

Auth errors

If you get auth errors, your connect.sid cookie has expired (~30 days). Grab a fresh one from your browser using the steps in Authentication.

Debugging with MCP Inspector```sh

npx @modelcontextprotocol/inspector uvx --from git+https://github.com/karbassi/loom-mcp.git loom-mcp
```

Contributing

Issues and pull requests are welcome on GitHub.

License

MIT

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Loom MCP 4fc5636a 怎么安装?

在客户端按下面的安装命令接入:uv run loom-mcp。装完在客户端配置里引用同一个包或端点。

Loom MCP 4fc5636a 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式是stdio · 本地进程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Loom MCP 4fc5636a 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Loom MCP 4fc5636a 需要 API Key 吗?

资料标注需要会话登录认证,使用前准备好对应凭据,并按最小权限配置。

Loom MCP 4fc5636a 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、VS Code;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Loom MCP 4fc5636a 能提供哪些能力?

资料显示约 59 个工具;能力标签包括 内容提取、Python、Claude、协作工具。具体工具清单以官方文档为准。

Loom MCP 4fc5636a 需要什么运行环境?

运行时为 Python;属于本地进程,需要在本机装好运行环境。

Loom MCP 4fc5636a 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。