QuickTube MCP Server

An MCP (Model Context Protocol) server for QuickTube. It lets AI agents such as Claude Code, Claude Desktop and Cursor fetch YouTube transcripts, summarize videos, find the key moments, generate quizzes, fact-check claims and more.

• Website: https://quicktube.ai
• API and MCP docs: https://quicktube.ai/api-docs
• Get an API key: https://quicktube.ai/settings?tab=api-keys

Tools

Setup

1. Get an API key

Create an API key at quicktube.ai/settings?tab=api-keys. Each tool needs a scope on the key:

generate_quiz and the YouTube data tools (get_video_metadata, search_videos, get_video_comments, get_related_videos) need no particular scope. Keys created on the settings page get the default scopes only, so the session and balance tools return 403 with them.

2. Add the server to your MCP client

The quickest route is uvx, which downloads and runs the server without a separate install step.

Claude Code

claude mcp add quicktube --env QUICKTUBE_API_KEY=qt_your_api_key_here -- uvx quicktube-mcp

Claude Desktop, Cursor and other clients (claude_desktop_config.json, .cursor/mcp.json, .mcp.json, ...):

{
  "mcpServers": {
    "quicktube": {
      "command": "uvx",
      "args": ["quicktube-mcp"],
      "env": {
        "QUICKTUBE_API_KEY": "qt_your_api_key_here"
      }
    }
  }
}

With pip instead of uvx

pip install quicktube-mcp

Then use "command": "quicktube-mcp" (no args) in the config above.

Environment variables

The server starts without QUICKTUBE_API_KEY so clients can list its tools, but every tool call fails until the key is set. QUICKTUBE_BASE_URL may be given with or without a trailing /api.

Long transcripts

get_transcript returns at most max_chars characters (80,000 by default, roughly 20k tokens) so a three-hour podcast doesn't flood the agent's context. When the result has "truncated": true, call it again with offset set to the returned next_offset. Each call costs a credit, so for a whole long video summarize_video, get_best_sections or chat_with_video are usually the better tools. get_session, list_sessions and batch_process_videos likewise leave out the timestamped transcript entries and cap the transcript text.

Errors

API errors come back to the agent as tool errors with the status and the server's reason, for example QuickTube API error 402 (Payment Required): Insufficient credits (needs 5 credits, balance 2). Buy credits at https://quicktube.ai/pricing, or a 429 with the Retry-After delay.

Running it directly

quicktube-mcp              # MCP over stdio (what MCP clients expect)
quicktube-mcp --help
quicktube-mcp --version
quicktube-mcp --transport streamable-http   # local HTTP transport

Example prompts

• "Get me the transcript of this YouTube video and summarize it."
• "What are the key moments in this podcast episode?"
• "Fact-check the claims made in this news video."
• "Generate a quiz from this educational video."
• "Is this video politically biased? Analyze it."
• "Process these 5 videos and give me summaries of each."

Development

pip install -e '.[test]'
pytest
quicktube-mcp --help

License

MIT