
Stream Clipper
io.github.lisn0由 GitHub 账号 lisn0 发布v0.1.0更新于 Nov 17, 2026
List local clips and prepare them for review or publishing.
发布方自己的服务:Stream Estate
概览
本地 stdio MCP 服务器,让助手浏览 Stream Clipper 保存的已完成 Twitch 切片,并可选地将切片保存为 PostFling 草稿。
- 可以做什么
- 列出本地切片目录中已完成的切片,最新的在前,含切片 ID、频道登录名、时间戳以及可选的时长、评分和触发字段
- 通过 limit 和 offset 分页浏览切片列表
- 可选地将选中的切片上传到已配置的 PostFling 工作区,并保存为映射目标的可编辑草稿
- 适用场景
- 当你已经在运行 Stream Clipper 监控程序从 Twitch 直播中切出片段,并希望助手在聊天客户端内查看、挑选或转交已完成的切片时使用。
- 运行要求
- 以本地进程运行,通过 npx(Node.js)或 uvx/PyPI 启动,并以参数传入切片目录的绝对路径;该目录中需已有完成的 MP4 文件及其 insights 附属文件。可选的草稿工具需要 PostFling 配置 JSON 路径,以及服务器环境中的 POSTFLING_EMAIL 和 POSTFLING_PASSWORD。Twitch 监控程序需单独运行,并需要 Twitch 凭据和视频工具。
安装
在 SourceWeft 中
- 打开 控制台中的 Stream Clipper,将其添加到工作区。
- 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
Stream Clipper
Automatically cut local clips from live Twitch streams using chat activity spikes. Stream Clipper monitors selected channels and saves candidate MP4s for review. Browse finished clips through the local MCP server or send selected clips to PostFling as drafts.
PyPI · npm · MCP Registry
Clip live Twitch streams · Browse local clips with MCP
Clip live Twitch streams
Requirements
- Python 3.10 or newer. The core pipeline uses only the Python standard library.
- A Twitch application client ID and secret for live-stream polling.
- yt-dlp and
FFmpeg and ffprobe on
PATHfor recording and export. The chat-only checks do not require video tools. - Optional dependencies include Tesseract with English data for
--fortnite-hudand faster-whisper with a cached Whisper Small model for speech-assisted options.
Install from PyPI
Install the CLI from PyPI in a Python virtual environment:
Use stream-clipping to run the monitor and stream-clipping-postfling to
preview or upload clips for draft review. From a source checkout, you can also
run python3 pipeline/live_monitor.py and python3 pipeline/push_to_postfling.py
with the same arguments.
Twitch use and authorization
The monitor reads stream status from Twitch Helix and chat from Twitch IRC. Live recording uses yt-dlp. Use it only where you have authorization to record the broadcast and retain the footage and chat under Twitch's Terms of Service. Obtain the necessary rights before sharing clips, including rights for any third-party material in the clip. Twitch also documents its official clip creation and retrieval API.
The monitor requires an explicit --channels list; there is no implicit channel
selection. Use channel login names separated by commas.
Configure and run
Create a local Twitch credential file (keep it outside source control):
Save it as secrets/twitch.env or another local path. Do not commit it. First
run a short dry-run to check Twitch polling and chat connectivity. This mode
does not record video:
After reviewing the dry-run output, start live recording and clipping:
Stop with Ctrl-C. The staging directory defaults to ./output/live. The
monitor keeps recordings and event logs there. Finished MP4s and their
*_insights.json sidecars go in clips/ under staging. The default disk
reserve pauses recording as free space falls. Recordings and exported clips
can accumulate across sessions, so set a retention policy for staging.
By default, the chat log stores event metadata such as time and message rate.
--save-chat-samples adds short raw chat excerpts. Restrict access to staging
and retain recordings, chat logs, and clips only as long as necessary.
For a bounded connectivity check, use --smoke SECONDS with exactly one
explicit channel and the credential file. That channel must be live. Smoke mode
checks Twitch polling and chat but does not record. Use an active chat:
Optional PostFling draft review
PostFling is optional. Its uploader defaults to review mode, which uploads
the selected clips and saves editable workspace drafts for a person to review.
It does not initiate publishing in that mode. Preview locally first with
--dry-run:
To save drafts, load POSTFLING_EMAIL and POSTFLING_PASSWORD into the
process environment from your secret store and omit --dry-run:
The default origin is http://127.0.0.1:8477 for local testing. For a remote
instance, set POSTFLING_URL to the HTTPS origin, such as
https://workspace.example, without an API path. The client rejects redirects.
Direct delivery is an explicit --mode direct action. Use it only when you
have the required rights and authorization for the content and destination.
Review each draft and destination before choosing direct delivery.
Browse local clips with MCP
The stdio server lets an MCP client browse completed live clips while the monitor runs separately. Use an absolute path to a clips directory writable only by you and the monitor. Create it or wait for the monitor to save a clip before connecting the client.
With Node.js and uvx on your MCP client's PATH, launch the server from npm:
The launcher uses the published PyPI package. Call list_clips to see completed
clips in that folder.
Run with uvx or install from PyPI
With uvx, the client can launch the PyPI package with its MCP dependencies:
If your client cannot find uvx on its PATH, use the absolute path to it.
You can also run python -m pip install 'stream-clipper[mcp]' in a virtual
environment and point the client at the installed stream-clipping-mcp
executable with --clips-dir and the absolute directory path as arguments.
Launch from a source checkout
Install the MCP Python SDK in a virtual environment for the client to launch.
Use absolute paths in your client's MCP server configuration.
MCP tools
list_clips shows the newest completed MP4 and insights pairs first. It returns
clip IDs, Twitch logins, Unix timestamps, and optional duration, score, and
trigger fields. Use limit (1–100, default 25) and offset (default 0) to page
through results.
To enable save_postfling_draft, append --postfling-config and the absolute
path to a JSON file to args. Use the same workspace and Twitch-login mapping
as the live monitor. Supply POSTFLING_EMAIL and POSTFLING_PASSWORD to the
server process through your client's secret environment or launch wrapper.
Call save_postfling_draft with a clip_id from list_clips. The tool uploads
that MP4 to the configured PostFling workspace and saves an editable draft for
the mapped destination. Review and publish through PostFling.
Client plugins
The Claude Code, Antigravity, Cursor, and Gemini CLI packages launch the local MCP
server. Put Node.js and uvx on the client's PATH. Set Clips directory to
the absolute path to a folder containing completed MP4 files and their
*_insights.json sidecars.
Open the plugin from this repository. In Claude Code, run claude --plugin-dir ./plugins/claude
from a checkout. In Antigravity, use agy plugin install .. In Cursor, use
Import from Disk in Customize. For Gemini CLI, install this repository with gemini extensions install https://github.com/lisn0/stream-clipper. Start Gemini CLI from your clips folder.
Ask the client to list the newest clips. The plugin browses completed clips
using list_clips. The Twitch monitor is started separately.
Tests
With FFmpeg and ffprobe installed, run the full suite from the repository root:
License
Project code is under the MIT License. The external tools and model weights retain their own licenses.
Maintainer releases
Set the same version in pyproject.toml and package.json. Build and test the
Python distributions and the npm archive from that source. Attach these three
files to the stable GitHub release before publishing it.
stream_clipper-X.Y.Z-py3-none-any.whlstream_clipper-X.Y.Z.tar.gzstream-clipper-X.Y.Z.tgz
Publishing the GitHub release with tag vX.Y.Z uploads the Python distributions
to PyPI first and the npm launcher second. Both registries require a trusted
publisher for publish-pypi.yml. The PyPI job uses the pypi environment.
To update the MCP Registry listing, match the versions in server.json to the
published npm package. Push the manifest, then use mcp-publisher publish server.json
from a machine authenticated to the official registry.
来源:README.md,提交 fb9e15d
工具
0版本历史
1- v0.1.0最新Nov 17, 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
其他 Stream 集成
Stream Estate
estate.stream官方发布者
远程 MCP 服务器,通过 Stream Estate API v2 让助手访问法国房地产房源、市场数据和房产提醒。

Deutschlandfunk — German public radio news: rolling news, topics, articles & live stream
由 GitHub 账号 mohammad-abbasi-sh 发布
只读 MCP 服务器,让助手获取德国之声 Deutschlandfunk 的滚动新闻、主题新闻、文章全文与直播流链接。

AuraStream
由 GitHub 账号 UnbiasedAlpha 发布
AuraStream MCP 通过一个 API 从 45 个以上来源获取历史行情、宏观、持仓与基本面时间序列,让助手直接取数并对比。

GenPark Reservoir Sampling Stream Auditor
由 GitHub 账号 alphaparkinc 发布
本地 MCP 服务器,提供无偏蓄水池抽样与分位数估计,面向 LLM 遥测等数据流。

pubmed-stream
由 GitHub 账号 melhzy 发布
检索 PubMed 与 PMC,下载开放全文并紧凑读取,梳理引用关系,并生成 Obsidian 知识图谱。