YouTube Transcriber

You deploy a small MCP server that exposes YouTube transcription capabilities as an MCP tool. It lets clients request a transcript, video title, thumbnail, and video ID for YouTube videos, all served through a scalable MCP endpoint you can host on Railway or run locally with uv. This guide walks you through using the server, installing it locally or remotely, and understanding how to extend it with more tools.

How to use

Once the MCP server is running, you can access the YouTube transcription tool via its MCP endpoint. Send a request to the server to transcribe a YouTube video by providing the video URL or ID. The server returns structured data including the video title, the full transcript, the thumbnail URL, and the video ID. You can run the server remotely on Railway or locally with uv and connect your MCP client to the endpoint.

How to install

Prerequisites you need before installing: a Railway account (Hobby plan recommended) and a GitHub account.

# Prepare your local machine for initial setup
# 1) Push this repo to GitHub
cd pocket-joe-mcp-toys
git init
git add .
git commit -m "Initial commit"
git remote add origin https://github.com/YOUR_USERNAME/pocket-joe-mcp-toys.git
git push -u origin main

# 2) Deploy to Railway
# - Go to Railway Dashboard
# - Click "New Project" → "Deploy from GitHub repo"
# - Select `pocket-joe-mcp-toys`
# - Railway auto-detects Python and deploys
# - Wait for deployment (takes ~2-3 minutes)

# 3) Get your endpoint
# - Click on your deployment
# - Copy the public URL: https://your-app.railway.app

Additional sections

Configuration and deployment details are important for reliable operation. The server is designed to be stateless and horizontally scalable, with a Python 3.12 runtime and the uv runtime managed via pyproject.toml. You can expose the MCP endpoint either remotely (HTTP) or locally (stdio) for development and testing.

Security: limit access to your Railway deployment or local environment as you would with any HTTP service. Do not expose the endpoint to untrusted clients in production unless you have authentication and authorization in place.

Cost and scale: on Railway Hobby plan you receive a monthly credit and potential for several hundred thousand requests, with auto-sleep after inactivity and options for a custom domain.

Extending with more tools: you can add additional pocket-joe policies as MCP tools by defining policy classes in your code, binding them to the AppContext, and creating FastMCP tools that invoke those policies. Then push to GitHub and redeploy; Railway will auto-redeploy.

Available tools

TranscribeYouTube

Transcribes a YouTube video and returns a structured result containing the video title, full transcript, thumbnail URL, and video ID.