mcp-youtube

This MCP server acts as a bridge between the YouTube API and AI assistants using the Model Context Protocol. It enables AI clients to request data related to YouTube videos, such as downloading closed captions, and can be extended with additional tools by editing the server code. The server is designed to be discovered by MCP clients (like Claude Desktop) and exposes its capabilities through a structured set of tools that map to YouTube-related actions. When you run the server, it becomes available as mcp-youtube for your MCP ecosystem to query and invoke.

You can interact with the server using Claude Desktop or the MCP inspector. The core capability highlighted in the README is the ability to download closed captions for a given video and to create summaries of the video content. Tools can be added by defining new ToolArgs classes and corresponding tool_runner implementations, allowing the server to grow with more YouTube data operations as needed. The inspector can be used to validate and debug tool availability and responses in a graphical UI, improving the reliability of tool usage in client applications.

Prerequisites:

• Install the uv tool (as noted in the README).

Step-by-step:

1. Install the YouTube MCP server via uv: uv tool install git+https://github.com/sparfenyuk/mcp-youtube

2. If you already have the server installed, you can upgrade or reinstall: uv tool upgrade --reinstall

3. If you need to uninstall the server: uv tool uninstall mcp-youtube

4. Verify installation by listing tools or running the server help command: uv sync uv run mcp-youtube --help

Prereqs overview:

• A working Python environment (as needed by the uv tool package).
• Access to YouTube API credentials if the server requires authenticated requests.

Tips and common issues:

• Ensure the full path to the mcp-youtube binary is used in client configurations if you encounter connection issues in Claude Desktop.
• When updating, use the provided uv upgrade command to avoid breaking changes.
• The server can be extended by adding new ToolArgs classes and implementing a corresponding tool_runner function; after changes, restart the server and refresh the client to expose new tools.
• For debugging, you can use the MCP inspector with: npx @modelcontextprotocol/inspector uv run mcp-youtube
• If you modify configuration files, ensure JSON formatting is valid to avoid startup failures.