Streams and playback
See which channels are live, start or stop a stream, and inspect or reorder what will play next.
Live Streaming Model Context Protocol (MCP)
Manage your Upstream live streams from the AI assistant you already use.
Upstream MCP is a live streaming MCP server for AI assistants. Check what is live, schedule broadcasts, organize media, tune playback, and prepare multistream destinations from Claude Code, Kimi Code, VS Code, or your own OpenAI integration.
Live streaming MCP, explained
A live streaming MCP server gives AI assistants tools for managing broadcasts.
Model Context Protocol (MCP) is an open standard that lets AI assistants discover and use tools. Once Upstream is connected, you can ask what is live, check whether media is ready, review a schedule, or propose a change without copying details between apps.
The Upstream MCP server turns those requests into authenticated actions through the Upstream live streaming API. It can check media processing, manage playback queues for 24/7 live streaming, update streams, configure automated live streaming schedules, and work with destinations for multistreaming. Confirmation behavior depends on your AI client and its settings.
Everyday live streaming tasks
Ask about the channel and make the change in the same conversation.
Use the Upstream MCP server to ask what is live, whether a video is ready, where a stream will publish, or what is next in the queue. Then use the relevant tool without hunting through separate screens.
Media and folders
Find videos, audio, images, and fonts; check processing; update details; and organize files into folders.
Schedules and destinations
Plan recurring broadcasts and manage where a stream publishes, including YouTube, Twitch, Kick, custom RTMP, and SRT.
Stream settings and organization
Review encoding, volume, recording, transitions, and video fit, then organize channels and campaigns with tags.
Complete broadcast workflows
Manage the whole broadcast, not just one setting.
Combine schedule, playback, media, and destination checks in one request. Ask for a plan first, then approve only the live streaming changes you want.
Schedule a 24/7 channel
Set scheduled starts and stops, repeat a broadcast, define continuous active and break periods, and preserve playback position between restarts.
Try asking
Inspect [stream name]. Propose a continuous schedule with 55 minutes live and a 5-minute break, preserve playback position, and wait for my approval before changing it.
Tune playback and output
Adjust resolution, frame rate, bitrate, recording, volume, loop or shuffle behavior, transitions, and how video fits the canvas.
Try asking
Show the current playback and output settings for [stream name]. Flag anything unusual and propose improvements without applying them.
Organize a growing media library
Search videos, audio, images, and fonts; check processing; update metadata; create folders; and organize multiple files together.
Try asking
Find processed media containing [keyword], show its current metadata and folders, then propose how to organize it. Do not move anything yet.
Prepare every destination
Create streams and manage YouTube, Twitch, Kick, custom RTMP, and SRT destinations, including whether they follow the stream schedule.
Try asking
Show every destination for [stream name] and which ones follow its schedule. Give me a preflight summary and wait for approval before changing anything.
Connect in three steps
Connect once, then start with a read-only check.
Create a dedicated API key, add the Upstream MCP server to your AI client, and verify that it can see the right live streaming account before asking it to make changes.
- 01
Create a dedicated API key
Open Profile → API Keys in Upstream Studio. Give the key a recognizable client name and save it when it is shown.
- 02
Add the Upstream MCP server
Choose your AI client below, add the Streamable HTTP endpoint, and store the API key in the client's secret settings.
- 03
Verify the connection
Ask the client to list your streams without making changes. If the client offers approval prompts, enable them before asking it to write or delete data.
Client setup
Choose the client you already use.
Claude Code, Kimi Code, VS Code, and other clients that support custom headers can connect today. ChatGPT requires OAuth, so direct ChatGPT connections are not available yet. Choose a client for exact steps, or use Other for the general setup.
Never paste a real API key into a shared file, public repository, support ticket, prompt, or directory submission.
VS Code
SupportedVS Code stores the server definition and asks for the secret value separately.
- 1. Run MCP: Open User Configuration.
- 2. Paste the configuration into
mcp.json. - 3. Enter your API key only when VS Code prompts.
mcp.json
{
"inputs": [
{
"type": "promptString",
"id": "upstream-api-key",
"description": "Upstream API key",
"password": true
}
],
"servers": {
"upstream": {
"type": "http",
"url": "https://studio.upstream.so/mcp",
"headers": {
"Authorization": "Bearer ${input:upstream-api-key}"
}
}
}
}Claude Code
SupportedAdd Upstream once at user scope, then inspect the connection from Claude Code.
- 1. Replace the placeholder locally.
- 2. Run the command in your terminal.
- 3. Run
/mcpto confirm the connection.
Terminal
claude mcp add --transport http --scope user upstream \
https://studio.upstream.so/mcp \
--header "Authorization: Bearer YOUR_API_TOKEN"Kimi Code
SupportedKimi Code supports Streamable HTTP servers with custom request headers.
- 1. Save this as
~/.kimi-code/mcp.json. - 2. Replace the placeholder only in that local file.
- 3. Start a new session and run
/mcp.
~/.kimi-code/mcp.json
{
"mcpServers": {
"upstream": {
"url": "https://studio.upstream.so/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_TOKEN"
}
}
}
}ChatGPT
OAuth requiredDirect connection is not available with Upstream's current API-key authentication.
Why there is no key to paste into ChatGPT
ChatGPT connects protected MCP servers through an OAuth sign-in flow. Upstream currently accepts a per-client API key instead, so we do not present an unsupported or unsafe workaround.
Available now: developers can connect from server-side code with OpenAI's Responses API and pass the key in the MCP tool's headers object. Keep it in your secret store.
Other MCP clients
Check compatibilityUse these details if your client supports remote Streamable HTTP servers and custom request headers.
- 1. Open your client's MCP or integrations settings.
- 2. Add a remote HTTP server using the details shown.
- 3. Save the key in the client's secret store, not a shared project file.
- 4. Reconnect, then confirm that the client discovers Upstream's tools.
Connection details
Transport: Streamable HTTP
Endpoint: https://studio.upstream.so/mcp
Header: Authorization: Bearer YOUR_API_TOKENThe setting names vary by client. If it cannot send a custom header, it cannot connect with Upstream's current API-key authentication.
Developer resources
Build directly with the same Upstream live streaming API.
The Upstream MCP tools use the customer API for streams, media, queues, folders, tags, and destinations. Browse the REST API reference or download a machine-readable specification for your own integration.
These resources are public and contain no customer data or API keys. Calling the API or MCP tools still requires authentication.
FAQ
Upstream MCP and live streaming MCP questions
What a live streaming MCP server is, what Upstream MCP can do, which clients work, and how to connect safely.
Upstream MCP is a remote live streaming MCP server. It connects an MCP-compatible AI assistant to authenticated tools for managing Upstream streams, schedules, playback queues, media, folders, tags, and multistream destinations.
Manage your next broadcast from the AI client you already use.
Create a dedicated API key, connect the Upstream live streaming MCP server, and start with a read-only check. Then schedule broadcasts, organize media, tune playback, and prepare destinations from the same conversation.