MCP server

Let an AI agent run your live streams.

Connect ChatGPT or Claude with your LiveReacting account, or Claude Code, Codex, Cursor and other AI agents with an API key. The agent builds a stream from a playlist, starts it, schedules it and reads the broadcast history afterwards.

https://mcp.livereacting.com/mcp

Streamable HTTP, POST only, no session. An API key goes in the Authorization header as a Bearer token. ChatGPT and Claude sign in with OAuth instead.

Get an API key Set up your client See the ChatGPT plugin
  • Included with every paid plan
  • 30 tools, with a read-only mode
  • Nothing to install

This page as Markdown

Setup

Connect your client.

One URL and one Authorization header for every client below. The key is the lr_ key the REST API takes, so a key you already have works here.

Let your coding agent set it up

Copy this message into your coding agent. The agent adds LiveReacting to its own settings, tells you where to put your API key and checks the connection.

Written for your agent to follow.

Use it in Claude Code, GitHub Copilot, Codex, Cursor, OpenCode, Antigravity, Gemini CLI, Kiro, Cline, Zed or another coding agent that can change its own settings.

Connect the LiveReacting MCP server to this agent.

1. Add a remote MCP server named "livereacting" to your own MCP settings, the way you normally add one (your config file or your "mcp add" command).
   URL: https://mcp.livereacting.com/mcp (Streamable HTTP)
   Header: Authorization: Bearer <my LiveReacting API key>
   Exact settings for Claude Code, Codex, Cursor, VS Code, Gemini CLI and more agents: https://www.livereacting.com/mcp.md
2. I create the API key at https://studio.livereacting.com/developer (every paid plan includes one). Keep it out of this chat and out of config files where you can: read it from the LIVEREACTING_API_KEY environment variable or your password prompt, and tell me how to set it. Never write the key into a file tracked by git or anywhere public.
3. Check the connection: call livereacting_list_projects and tell me what it returned. If the server loads only after a restart or a new session, tell me exactly what to do, and to ask you "List my LiveReacting projects" afterwards.

If you cannot change your own settings (for example, you are a chat app), say so and send me to https://www.livereacting.com/mcp.

If you have no key yet, create one in the Developer section of the LiveReacting Studio. Set it as the LIVEREACTING_API_KEY environment variable and restart your client. VS Code asks for the key instead.

Claude Code

In .mcp.json at the root of your project. The key stays out of the file, so it is safe to commit.

{
  "mcpServers": {
    "livereacting": {
      "type": "http",
      "url": "https://mcp.livereacting.com/mcp",
      "headers": { "Authorization": "Bearer ${LIVEREACTING_API_KEY}" }
    }
  }
}

Codex CLI

In ~/.codex/config.toml.

[mcp_servers.livereacting]
url = "https://mcp.livereacting.com/mcp"
bearer_token_env_var = "LIVEREACTING_API_KEY"

# Then, in your shell profile:
# export LIVEREACTING_API_KEY=lr_...

Cursor

In ~/.cursor/mcp.json.

{
  "mcpServers": {
    "livereacting": {
      "url": "https://mcp.livereacting.com/mcp",
      "headers": { "Authorization": "Bearer ${env:LIVEREACTING_API_KEY}" }
    }
  }
}
Add to Cursor

VS Code (GitHub Copilot)

In .vscode/mcp.json. The inputs block makes VS Code ask for the key instead of storing it in the repository.

{
  "inputs": [
    {
      "id": "livereacting-key",
      "type": "promptString",
      "description": "LiveReacting API key",
      "password": true
    }
  ],
  "servers": {
    "livereacting": {
      "type": "http",
      "url": "https://mcp.livereacting.com/mcp",
      "headers": { "Authorization": "Bearer ${input:livereacting-key}" }
    }
  }
}
Add to VS Code

Gemini CLI

In ~/.gemini/settings.json.

{
  "mcpServers": {
    "livereacting": {
      "httpUrl": "https://mcp.livereacting.com/mcp",
      "headers": { "Authorization": "Bearer ${LIVEREACTING_API_KEY}" }
    }
  }
}

More coding agents

Agent Settings file API key
GitHub Copilot CLI ~/.copilot/mcp-config.json"mcpServers": { "type": "http", "url", "headers" } Typed into the file. Its docs show no way to read it from your environment.
OpenCode ~/.config/opencode/opencode.json"mcp": { "type": "remote", "url", "headers" } Read from your environment with {env:LIVEREACTING_API_KEY} in the header.
Antigravity ~/.gemini/config/mcp_config.json"mcpServers": { "serverUrl", "headers" } Typed into the file. Its docs show no way to read it from your environment.
Kiro ~/.kiro/settings/mcp.json"mcpServers": { "url", "headers" } Read from your environment with ${LIVEREACTING_API_KEY} in the header, once you approve the variable in Kiro.
Cline cline_mcp_settings.json"mcpServers": { "type": "streamableHttp", "url", "headers" } Typed into the file. Its docs show no way to read it from your environment.
Zed settings.json"context_servers": { "url", "headers" } Typed into the file. Its docs show no way to read it from your environment.

Any client that supports Streamable HTTP and a Bearer header takes the same URL and the same header.

Test safely first

Add ?read_only=1 to the URL and the client is offered only the tools that read. Nothing the agent calls can change a project or a broadcast.

Tools

What the agent can call.

Every name starts with livereacting_, so one search finds the whole server in a client that loads tools on demand.

Tool What it does Access
Account The destinations you have already connected.
livereacting_list_destinations Your connected YouTube, Facebook, Twitch, LinkedIn, X and your own servers Read only
livereacting_validate_destinations Check a YouTube, Facebook or Twitch login still works before you rely on it. Custom destinations are not contacted Read only
Editor Projects, scenes and the video and audio playlists inside them.
livereacting_list_projects Your projects, newest first Read only
livereacting_get_project One project in full: format, duration, destinations and stream status Read only
livereacting_create_project Create an empty project, the first step of building a stream Writes
livereacting_update_project Change a project's settings Writes
livereacting_get_scenes The scenes of a project, with a summary of what is in each one Read only
livereacting_manage_scene Create, rename or delete a scene Destructive
livereacting_activate_scene Switch the scene that goes out on air, including while the stream runs Writes
livereacting_create_scene_with_playlist Add a scene and fill it with a video playlist in one call Writes
livereacting_add_layer Add one video playlist, audio playlist or single video to a scene that already exists Writes
livereacting_get_layer Read one playlist and the videos in it, fifty at a time Read only
livereacting_set_playlist Replace everything in a video or audio playlist Destructive
Streams Going live, scheduling, and watching a broadcast while it runs.
livereacting_start_stream Go live now Destructive
livereacting_stop_stream End the broadcast that is running Destructive
livereacting_get_stream_status What the stream is doing right now, and the state of each destination Read only
livereacting_schedule_stream Go live at one or more times in the next 28 days Writes
livereacting_cancel_schedules Cancel one scheduled broadcast, or every schedule on a project Destructive
livereacting_sync_stream Push the edits you just made to the stream that is running Writes
livereacting_list_broadcasts Past broadcasts: when each ran, for how long, and its peak viewers Read only
Media The video library the playlists are built from.
livereacting_list_media The files and folders in your library Read only
livereacting_get_media_file One file, with its technical detail when you ask for it Read only
livereacting_import_media Bring in one file from a Google Drive, Dropbox or YouTube URL Writes
livereacting_import_from_url Bring in one video, audio or image file from a direct link on a public site Writes
livereacting_get_import_status How far an import has got, and whether the file can be streamed yet Read only
livereacting_upload_file Start an upload from your computer. The agent sends the file with the commands it gets back Writes
livereacting_complete_upload Finish the upload and add the file to your library Writes
livereacting_encode_media Queue files for encoding so they can be streamed Writes
livereacting_get_encode_status How far the encoding has got Read only
Analytics The numbers a finished broadcast produced.
livereacting_get_stream_analytics YouTube's broadcast numbers on every paid plan, and per-video playlist numbers by request Read only

Each tool's input schema comes from tools/list once you connect, and its description says what to call next and how long to wait between polls. Read them there; do not infer them from the names above.

Add ?toolsets= with any of account, editor, streams, media and analytics, comma-separated, and the client is offered only those groups. The server does the filtering, so any client can use it, and it refuses a name it does not recognize.

Example prompts

What you ask for, and what the agent does.

Say what the stream should do, in plain words. The agent picks the tools from the table above and calls them in the order the job needs.

  • "Here are three Google Drive video links. Build a 24/7 live stream on my YouTube channel from them and start it."

    It imports the three files, waits until they can be streamed, fills a looping playlist and goes live.

  • "Schedule last week's webinar recording to go out live on Friday at 3pm."

    It finds the recording in your library, builds a project that plays it once, and schedules that one start time.

  • "We are live. Switch to the interval scene."

    It activates the interval scene and pushes the change to the running broadcast. The stream stays live.

  • "How did last week's broadcasts do?"

    It goes through your projects, lists the broadcasts each one ran with their peak viewers, then reads YouTube's numbers for them.

Limits

What to decide before you connect an agent.

  • Paid plans only

    The Free plan and a paused subscription do not include API access. With an API key, every request is refused with 403 API_ACCESS_REQUIRES_PAID_PLAN. Through ChatGPT or Claude the account can connect, and each tool call returns that error until the plan includes API access.

  • The key acts as your account

    Anything the key can reach, the agent can reach. Create it in the Developer section of the Studio, and revoke it there the moment you want the agent to stop.

  • Give the agent fewer tools

    Add ?read_only=1 for the tools that read, or ?toolsets=streams,media for the groups you name. The server refuses an unrecognized value rather than ignoring it.

  • What it cannot do yet

    It cannot import a YouTube playlist, or build a text, image or interactive overlay. You build those in the Studio, and the agent can still run a stream that has them.

Rules the agent should know

  • Stay under 150 tool calls a minute. Counted per account, and the initialize and tools/list handshake is free. A single call over the limit comes back as HTTP 200 with a tool error that says how long to wait, so the model reads it rather than a client retrying a 429 on its own. A batch is refused with a real 429, and a batch may carry at most 20 messages.
  • set_playlist replaces the whole list. It is not an append. To add one video, read every page of get_layer first, then send the complete list back.
  • Check before you retry a write. After a timeout, ask get_stream_status, or the list tool for what you were creating, whether the write went through before you call it again. A second create_project makes a second project, and a second add_layer adds a second layer.
  • The access labels are hints, not locks. Every tool that writes or deletes carries the annotation, and a client can use it to ask you before calling. The protocol does not make it ask, so check your client's own setting.
  • A 24/7 stream needs somewhere that allows one. Facebook caps a Facebook-only stream at eight hours, so a 24/7 one needs YouTube, your own server, or a second destination beside Facebook.
  • Protocol revisions 2024-10-07 through 2025-11-25. A client pinned to 2026-07-28 cannot connect, and that is the first thing to check when one will not.

Questions

Can the agent delete something?

Yes. It can delete a scene, replace a whole playlist, stop a broadcast and cancel a schedule, and the table above marks each of those tools Destructive. With ?read_only=1 on the URL the server offers none of them.

Which AI assistants work?

ChatGPT, through the LiveReacting plugin: open it from our ChatGPT page, install it and sign in to your LiveReacting account. In Claude, open Customize, then Connectors, choose Add custom connector, enter the server URL above and sign in the same way. In Mistral Vibe, add a Custom MCP Connector with the same URL, choose HTTP Bearer Token and paste your API key. Coding agents such as Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI set themselves up from the setup message above, and each has a configuration on this page. The Gemini app and Microsoft Copilot cannot connect to LiveReacting yet.

Can the agent upload a video from my computer?

Yes, when the agent can run commands on your computer. It asks LiveReacting for upload links, sends the file with curl, and the file joins your library. Each file can be as large as your plan allows. An agent without a terminal can import one file at a time from a direct public link, or from Google Drive, Dropbox or YouTube.

Where is the full reference?

Every tool, every input, every response field and every error code is in the MCP reference. The REST API it sits on is documented beside it.

Start with a free account.

Build your stream in the studio on the Free plan, which does not expire. Move to any paid plan when you are ready to hand it to an agent, and the API key comes with it.

  • No credit card
  • 4.8 from 633 reviews

Want the numbers first? See plans and pricing