LiveReacting MCP server

Run live video streams from an AI agent. Build a channel from a video playlist, start it, schedule it, watch it while it runs, and read the numbers afterwards — from Claude Code, Codex, Cursor, or anything else that speaks MCP.

https://mcp.livereacting.com/mcp

Streamable HTTP. Stateless. Authenticated with your LiveReacting API key.

Full reference, with every tool, input, response field, limit and error code: developers.livereacting.com/mcp (REFERENCE.md in this repo).

Get a key

API access is included with every paid plan. Create a key in the Developer section of the LiveReacting Studio. Keys start with lr_.

With an API key, an account on the Free plan gets 403 API_ACCESS_REQUIRES_PAID_PLAN on every call. With OAuth (ChatGPT, Claude), the account can connect, and each tool call returns an error that explains the plan limit. The tools start working as soon as the plan includes API access, without connecting again.

Setup

Create the key first, put it in your shell profile as LIVEREACTING_API_KEY, then paste the config below.

Every example that cannot read that variable uses lr_… as a placeholder. Do not paste a real key into a file you commit. Where a harness supports an environment variable or a prompt, the block uses it.

Claude Code

In .mcp.json. Claude Code expands ${VAR} in a remote server's headers (docs), so the key stays out of the file:

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

claude mcp add --transport http livereacting https://mcp.livereacting.com/mcp --header "Authorization: Bearer lr_…" also works, but it writes the key itself into the stored config.

Codex CLI

In ~/.codex/config.toml:

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

Then export LIVEREACTING_API_KEY=lr_… in your shell profile. Prefer this over http_headers, which puts the key in the config file.

Cursor

In ~/.cursor/mcp.json. Cursor resolves ${env:NAME} in headers (docs):

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

Or install the LiveReacting plugin from the Cursor Marketplace and set LIVEREACTING_API_KEY under Plugins → Configure. The plugin is .cursor-plugin/plugin.json and mcp.json in this repo.

VS Code (GitHub Copilot)

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

{
  "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}" }
    }
  }
}

Gemini CLI

Install the extension. Gemini CLI asks for the key once and keeps it in the system keychain:

gemini extensions install https://github.com/LiveReacting/livereacting-mcp

Or, without the extension, in ~/.gemini/settings.json. Gemini CLI expands $VAR and ${VAR} in every string of the file before it validates it, headers included (docs), so the key stays out of it:

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

Kiro

Install this repo as a power: Powers panel → Add Custom Power → Import power from GitHub, then paste https://github.com/LiveReacting/livereacting-mcp. The power reads LIVEREACTING_API_KEY from your environment (mcp.json).

Zed

In your Zed settings.json, under context_servers, add an HTTP server pointing at the URL with an Authorization header.

Goose

Add a Remote Extension (Streamable HTTP) in Goose's settings, URL https://mcp.livereacting.com/mcp, with a custom header Authorization: Bearer lr_….

Compatibility

Protocol revisions: 2024-10-07 through 2025-11-25. If your client reports a version error, that is the first thing to check. 2026-07-28 is not supported yet — no published TypeScript SDK speaks it.

Tools

30 tools in five toolsets. Every name is prefixed livereacting_, so one search in a client that defers tool loading finds the whole server.

account

editor

streams

media

analytics

Selecting fewer tools

Both are query parameters on the URL, so no client support is needed.

https://mcp.livereacting.com/mcp?toolsets=streams,media
https://mcp.livereacting.com/mcp?read_only=1

read_only=1 drops every tool that can change anything. Useful for a monitoring agent you do not want touching a live broadcast.

Limits, stated honestly

An agent that promises any of these is lying to your customer.

• You can build video and audio playlists, and nothing else. Text, image and interactive layers are readable through the API but can only be created or edited in the LiveReacting Studio.
• Operating is not limited that way. A stream whose overlays were built in the Studio can still be started, stopped, scheduled, synced and monitored from here. The line is author versus operate.
• Uploading a local file needs a shell or HTTP client. upload_file returns signed part URLs and ready curl commands; the agent sends the bytes itself, then calls complete_upload.
• Every plan has a maximum stream length. A stream stops when it reaches it, even with a manual duration. A 24/7 channel needs a plan without a length limit, such as the 24/7 plan.
• A Facebook-only 24/7 channel is not possible. Facebook-only streams get a fixed 8-hour maximum.
• Thumbnails and live announcements are unsupported.
• Playlist analytics is available on request. YouTube analytics is included with every paid plan, and needs the channel to have granted analytics access. Ask hello@livereacting.com for playlist analytics.

Rate limits

150 requests per minute per account, shared across your API keys.

REST and MCP are counted separately, so the two surfaces get 150 each rather than sharing one budget. Do not plan around that: it is an implementation detail of how the two are metered, not a promise.

A 429 comes back as a tool error that says the number and how long to wait, because the model is the one reading it. The polling tools — get_stream_status, get_import_status, get_encode_status — each document a minimum interval. Poll tighter than that and you spend your own budget on nothing.

Things worth knowing before your agent surprises you

set_playlist replaces the whole playlist. It is not an append. To add one video, read the layer first, then send the complete list back. The tool tells you when a write shrinks the playlist, but it will still do what you asked.

Starting and stopping a stream is not idempotent. If a call times out, ask get_stream_status what happened. Do not retry — a retry can start a second attempt.

Docs

Full MCP reference: developers.livereacting.com/mcp Full REST API reference: developers.livereacting.com Support: hello@livereacting.com Privacy policy: livereacting.com/privacy-policy