The Mux MCP (Model Context Protocol) Server brings Mux's Video and Data platform capabilities directly to your AI tools. Once set up, you can upload videos, manage live streams, analyze video performance, and access practically all of Mux's video infrastructure through natural language prompts in supported AI clients.

This guide walks you through how the Mux MCP server works and how to connect it to various AI clients. The quickest way to get started is the remote server hosted at https://mcp.mux.com.

How it works: Code Mode

The Mux MCP server is built on the "Code Mode" tool scheme. Rather than exposing hundreds of individual endpoint tools, it exposes just two tools to your agent:

• search_docs for querying documentation about Mux's API and SDKs.
• execute where your agent writes TypeScript against the full Mux SDK, @mux/ts. Anything the code returns or prints is handed back to the agent as the result of the tool call.

On the hosted server, each execute call runs in a fresh, isolated Mux-owned sandbox with network access limited to Mux API hosts.

Because the agent writes and runs real SDK code, it can perform complex, multi-step tasks deterministically and repeatably while keeping the model's context window small. The full surface of the Mux SDK is reachable this way, including:

• Mux Video: assets, direct uploads, playback IDs, tracks, live streams, and delivery usage
• Mux Data: metrics, dimensions, filters, real-time data, video views, and exports
• Mux Robots: jobs, workflows, and directives
• System: signing keys and account utilities (such as whoami)
• Webhooks: verifying and unwrapping webhook payloads

Because execute exposes the whole SDK, Mux Robots is fully usable through the MCP server, including directives.

Prompt examples

Video Management

• Using the Mux tool, create a webpage where I can upload a video to Mux
• Give me the playback URL for the most recently uploaded video to my Mux account, use Mux MCP
• List all my video assets and their current status (using the Mux MCP tool)
• With the Mux tool: Show me recent video uploads
• Using Mux MCP, generate a subtitles track for asset ID: ASSET_ID

Mux Robots

• Using the Mux MCP, run a Robots job to generate chapters for asset ID: ASSET_ID
• With the Mux tool, summarize my most recently uploaded video and suggest tags for it

Mux Data Analytics and Performance

• Using the Mux MCP, tell me the best performing country for video streaming over the last month
• Show me video performance metrics for the last week using the Mux tool
• With the Mux tool: what are the top performing videos by view count?
• Using Mux, which countries have the highest video engagement?
• What are the most common video errors in my account (use the Mux MCP)?
• Show me breakdown values for video quality metrics using the Mux MCP tool
• List all available data dimensions I can filter by, use the Mux MCP to answer this prompt

Prerequisites

Before utilizing the Mux MCP Server, make sure you meet the following prerequisites:

• A Mux account (sign up at mux.com if you don't have one)
• Claude Desktop, Cursor, or any other client that supports remote MCP servers, installed and updated to the latest version

Configuring the Mux MCP Server

Mux's MCP server is hosted at https://mcp.mux.com. To configure it in your client, add an MCP server — sometimes called a "connector" (Claude/Claude Code/ChatGPT), an "extension" (Goose), or simply an MCP Server (VS Code) — and enter the URL https://mcp.mux.com as the location.

Every request to the server, including the initial connection, requires credentials. Clients that support OAuth handle this automatically, so there's no need to grab Access Token information from the Dashboard. Your client prompts you to:

• Log in to dashboard.mux.com via whatever means you normally use (this is skipped if you're already logged in)
• Choose which environment you want to authorize this connection for

When you're already logged in, your experience will look something like this:

Once you've logged in and chosen an environment, you're good to go!

Clients that don't support OAuth

If your client can't do OAuth, you'll need a Mux access token, which you can create under Settings → Access Tokens in the Mux Dashboard. The simplest option is to run the server locally, which takes the token ID and secret directly as the MUX_TOKEN_ID and MUX_TOKEN_SECRET environment variables.

To use the hosted server instead, send the token as Basic credentials in the Authorization header: join the ID and secret as TOKEN_ID:TOKEN_SECRET and base64-encode the result. For clients with a configuration JSON, that looks something like this:

{
  "mcpServers": {
    "mux": {
      "url": "https://mcp.mux.com",
      "headers": {
        "Authorization": "Basic <base64 of TOKEN_ID:TOKEN_SECRET>"
      }
    }
  }
}

The access token determines which environment the server acts on and what it's allowed to do, so treat this configuration like any other secret.

A note on remote MCP support

These days, most LLM clients directly support remote MCP servers (rather than locally installed ones), so you shouldn't have much trouble getting set up. That said, there are still some clients (particularly older versions) that don't have built-in remote MCP support. For those situations, you have two options:

1. You can install the MCP server locally, which also lets you restrict the tools and SDK methods the server is allowed to run.
2. You can use mcp-remote, which brings remote MCP support to practically any LLM client (and may perform better than a client's built-in remote MCP support, depending on the client).

Having trouble?

If you run into issues or have questions:

• Check the Model Context Protocol documentation for general MCP setup guidance
• Review Claude's MCP documentation for Claude-specific configuration
• Visit our API reference for detailed endpoint documentation
• Contact support: mux.com/support