Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者提供的说明(英文原文)
Manage 24/7 live streams, media, playback queues, schedules, and multistreaming from AI assistants.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
适用于网页版、桌面版和手机版 Claude:
https://studio.upstream.so/mcp在终端中运行这条命令
claude mcp add --transport http upstream https://studio.upstream.so/mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"upstream": {
"url": "https://studio.upstream.so/mcp"
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"upstream": {
"type": "http",
"url": "https://studio.upstream.so/mcp"
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
ChatGPT 只能连接互联网上的服务器:
https://studio.upstream.so/mcp菜单名称可能因套餐和版本不同而略有差异。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
我们还没有读取这个服务器的工具列表;我们会轮流检查各个端点。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月6日
注册名称已通过域名 upstream.so 验证(注册中心会核实域名所有权)
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
尚未检查;我们会轮流检查各个端点
尚未检查;我们会轮流检查各个端点
尚未检查;我们会在检查端点时读取工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在作者的云端运行。通过互联网连接,无需安装,但你发送的内容会经过作者的服务器。
GitHub 仓库最后更新于–,最新版本(1.0.2)于34天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,所以接受远程 MCP 服务器的应用也可以使用,例如开启开发者模式的 ChatGPT。
同一分类中有:Spec Driven Development, Remote Desktop Commander, yontu
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Türkçe yazı denetimi: TDK kuralları, yapay zekâ izleri, metin türüne göre üslup.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
The official Model Context Protocol server for Upstream.so, the cloud live-streaming platform. Point an AI assistant at one URL and it can run your channels: start and stop streams, reorder playback queues, search your media library, plan schedules, and manage multistream destinations.
It is a remote server — nothing to install, no npm package, no local process. It mirrors the public Upstream API one-to-one, so every tool runs the same validation, ownership rules and rate limits as the REST endpoint behind it.
┌───────────────┐ MCP (Streamable HTTP) ┌────────────────────┐
│ your AI │ ────────────────────────► │ studio.upstream.so │ │ (Claude, ...) │ ◄──── tool results ────── │ /mcp │ └───────────────┘ └─────────┬──────────┘ │ same code path ▼ ┌────────────────────┐ │ Upstream API v1 │ └────────────────────┘
| URL | https://studio.upstream.so/mcp |
| Transport | Streamable HTTP |
| Auth | OAuth (recommended); API key fallback |
Connect with OAuth when your client supports it. Add the server URL, sign in directly on studio.upstream.so, and review the requested access. Your AI client receives an access token, never your Upstream password. Remove its access under Profile → Connected apps.
For clients with custom request headers, a Personal Access Token remains available under Profile → API Keys. Create a dedicated key for each client and store it in the client's secret settings. Existing API-key connections keep working.
Both methods grant access to the supported MCP tools for your account, including actions that change or delete data. OAuth's mcp:use scope is not a read-only mode. Enable your client's approval prompts and begin with a read-only request.
The legacy
X-Upstream-Api-Keyheader still works. New API-key connections should sendAuthorization: Bearer.
https://studio.upstream.so/mcp. Choose OAuth authentication when prompted.This uses a custom MCP connection. It does not require a published ChatGPT app. See OpenAI's connection guide for current availability and settings.
https://studio.upstream.so/mcp.Custom connectors depend on your plan. Team and Enterprise owners must first add the connector for their organization. See Claude's custom connector guide.
claude mcp add --transport http --scope user upstream \
https://studio.upstream.so/mcp
Open Claude Code, run /mcp, and select Upstream to authenticate. Sign in on Upstream and approve access, then run /mcp again to check the connection. See the Claude Code MCP guide.
Run MCP: Open User Configuration from the Command Palette and add Upstream to mcp.json, keeping any other servers already configured:
{
"servers": { "upstream": { "type": "http", "url": "https://studio.upstream.so/mcp" } } }
Start Upstream from the server configuration and follow the authentication prompt. Sign in on Upstream and approve access. VS Code discovers the OAuth settings automatically; no API key or client secret is needed. See the VS Code MCP guide.
Add Upstream to ~/.kimi-code/mcp.json, keeping any other servers already configured:
{
"mcpServers": { "upstream": { "url": "https://studio.upstream.so/mcp" } } }
Start a new Kimi Code session and run /mcp-config login upstream. Sign in on Upstream and approve access, then run /mcp to check the connection. You can also add the server interactively with /mcp-config. See the Kimi Code MCP guide.
Existing API-key connections keep working. To switch, remove the saved Authorization header or API-key setting from the Upstream server configuration, then reconnect and follow your client's sign-in steps above. A configured API-key header can prevent the client from starting OAuth.
Scripts and clients without OAuth can use a dedicated key from Profile → API Keys. Use the same endpoint and store this header in the client's secret settings:
Authorization: Bearer YOUR_UPSTREAM_TOKEN
Replace the placeholder with your key only in those settings. Never paste a real key into a prompt, shared project configuration, support ticket, or directory submission.
Other clients need remote Streamable HTTP and OAuth or custom request headers. OAuth callback compatibility depends on the client.
Verify the connection with a read-only call: ask the assistant to list your streams.
Tools use the same account ownership, validation, and rate limits as the public API.
| Tool | What it does |
|---|---|
list_streams | List the streams owned by the account, newest first, paginated. |
get_stream | Get one stream by UUID: settings, status, tags, destinations. |
create_stream | Create a stream — name, platform (youtube, twitch, kick, custom, customsrt), stream key. |
update_stream | Partially update a stream; only provided fields change. |
delete_stream | Permanently delete a stream and its configuration. |
start_stream | Start broadcasting to the stream's platform. |
stop_stream | Stop a running stream. This interrupts a live broadcast for viewers. |
Scheduling is part of a stream rather than a separate tool: create_stream and update_stream both take a schedule object covering start and end times, repeats, and continuous mode with its own run and break durations.
Each stream owns queues by type: video, audio, audio-secondary, external-videos.
| Tool | What it does |
|---|---|
get_stream_queue | Get the ordered items of one queue. |
add_to_stream_queue | Append media UUIDs to a queue — or {url, name} objects for external-videos. |
remove_from_stream_queue | Remove one item. The media file itself is not deleted. |
reorder_stream_queue | Replace a queue's playback order with a full ordered list of ids. |
| Tool | What it does |
|---|---|
list_media | List and search the library, paginated. Filter by type, name or folder. |
get_media | Get one file, including its processing state. |
update_media | Update name and audio tags (title, artist, album, year, genre). |
delete_media | Permanently delete a file. It also disappears from any queue referencing it. |
create_upload_ticket | Mint a TUS upload ticket so the client can upload files itself. |
revoke_upload_tickets | Revoke active upload tickets for the account. |
A freshly uploaded file is *processing* until get_media reports is_processing: false. It can only be queued after that.
| Tool | What it does |
|---|---|
list_folders | List folders, optionally filtered by parent. |
get_folder | Get one folder by UUID. |
create_folder | Create a folder, optionally inside a parent. |
update_folder | Rename a folder, or move it under another parent. |
delete_folder | Delete a folder and its subfolders. Contained files move to the library root. |
move_files_to_folder | Move up to 100 files into a folder, or back to the root. |
| Tool | What it does |
|---|---|
list_destinations | List the extra platforms a stream is restreamed to. |
add_destination | Add a destination — name, platform, stream key. |
update_destination | Partially update a destination. |
remove_destination | Remove a destination from the stream. |
| Tool | What it does |
|---|---|
list_tags | List the account's tags. |
create_tag | Create a tag. Names are unique per account. |
update_tag | Update a tag's name, colour or icon. |
delete_tag | Delete a tag. Streams keep working; they just lose it. |
get_account | Get the authenticated account: id, name, email, plan and limits. |
MCP carries JSON, not file bytes, so create_upload_ticket hands the upload back to the client: it returns a TUS endpoint and a 24-hour token, and the assistant's own shell moves the bytes.
Reuse one ticket per batch when practical. Minting a new ticket renews the same family without revoking in-flight uploads. Use revoke_upload_tickets to invalidate active tickets. Runnable examples in several languages live in upstream-upload-examples.
Tool calls share the REST API's budget of 60 requests/minute per account. On HTTP 429, back off.
Resources belong to the token's account only: a foreign id returns 403, an unknown id returns 404. Destructive tools carry MCP annotations, so clients that render confirmations will ask before deleting or stopping something.
Upstream runs live video in the cloud, without OBS or a machine of your own left switched on:
MIT — see LICENSE.