YouTube Transcript MCP Server: Setup Guide for Claude and Other AI Tools

The Model Context Protocol standardizes how an AI client discovers and calls external tools. Under the hood it's JSON-RPC 2.0 - the client asks the server what tools exist and their input schemas, then invokes them by name with structured arguments and gets a structured result back, all inside the same conversation turn.

MCP servers run over one of two transports: stdio, where the client launches the server as a local subprocess (typical for tools that need local filesystem access), or Streamable HTTP, where the server is remote and reachable over HTTPS by any number of clients at once - which is what this server uses. (An older HTTP+SSE transport exists in some documentation but is deprecated in favor of Streamable HTTP.) Practically, this means you never run anything locally to use this server - just point your client at the URL.

The paginated tools (search, list-channel, list-playlist) all accept a continuation token from a previous response to fetch the next page, rather than an offset/limit pair.

Sign up from the dashboard and create an API key - 100 credits are included free, no card required. This key is the bearer token the MCP server authenticates with for most clients.

For Claude Desktop, add this to claude_desktop_config.json - on Mac that's ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "getyoutubetranscript": {
      "url": "https://getyoutubetranscript.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_live_..."
      }
    }
  }
}

Replace sk_live_... with your own API key. Restart the client fully and the 7 tools above appear automatically - most clients only read this config on startup, not on a hot reload.

For Claude Code, skip the JSON file entirely and register the server from the CLI instead:

claude mcp add --transport http getyoutubetranscript \
  https://getyoutubetranscript.com/api/mcp \
  --header "Authorization: Bearer sk_live_..."

Claude Code stores this per-project by default (-s local); pass -s user to make it available across every project instead.

Any client that supports a remote MCP server over Streamable HTTP with custom headers uses the same shape of config - just the surrounding file and field names differ (e.g. Cursor's .cursor/mcp.json, Windsurf's MCP settings panel). If a client only supports OAuth-based servers rather than a static bearer header, use the OAuth flow described next instead.

For clients that authenticate MCP servers via OAuth rather than a pasted API key (for example, ChatGPT's custom connectors), this server also supports OAuth 2.1 with Dynamic Client Registration - point the client at the same server URL and it discovers the OAuth metadata and walks you through an authorization flow instead of asking for a bearer token directly.

Once connected, natural prompts are enough - the model decides which tool to call:

• "Get the transcript for youtube.com/watch?v=dQw4w9WgXcQ and summarize it in 3 bullets."
• "Search YouTube for 'Next.js server actions' and list the top 5 results."
• "List every video @mkbhd has uploaded and find ones mentioning 'battery'."