YouTube Content Automation

A local AI-powered YouTube content management system. Claude Code acts as the AI brain; an MCP server handles all data operations against Google Sheets, Docs, Drive, and web research APIs.

You (conversation) → Claude Code (AI reasoning) → MCP Tools (data layer) → Google APIs

Prerequisites

• Python 3.10+
• Claude Code CLI installed
• A Google Cloud project with the Sheets, Docs, and Drive APIs enabled
• A Google Sheet to use as your content database

Installation

git clone <repo-url>
cd youtube-mcp

python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate

pip install -r requirements.txt
pip install -e .                   # installs the `content` CLI command

Google Cloud Setup

1. Go to Google Cloud Console and create or select a project.
2. Enable these APIs: Google Sheets API, Google Docs API, Google Drive API.
3. Under APIs & Services → Credentials, create an OAuth 2.0 Client ID (Desktop app type).
4. Download the JSON file and save it as client_secret.json in the project root.

Environment Configuration

cp .env.example .env

Open .env and fill in:

Variable	Where to find it
GOOGLE_CLIENT_ID	OAuth client ID from Cloud Console
GOOGLE_CLIENT_SECRET	OAuth client secret from Cloud Console
GOOGLE_REFRESH_TOKEN	Generated during the auth step below
GOOGLE_SHEET_ID	The ID in your Sheet URL: spreadsheets/d/<ID>/edit
GOOGLE_DRIVE_FOLDER	The ID in your Drive folder URL: drive/folders/<ID>
BRAVE_SEARCH_API_KEY	Optional — Brave Search API for web research
REDDIT_CLIENT_ID	Optional — Reddit API app credentials for Reddit research
REDDIT_CLIENT_SECRET	Optional — Reddit API app credentials

One-Time Google Auth

Run the OAuth flow once to generate a refresh token:

python -m commands.cli auth client_secret.json

A browser window will open asking you to sign in. After approval, the terminal prints your GOOGLE_REFRESH_TOKEN — copy it into .env. The token is also cached locally in config/token.json and auto-refreshed on subsequent runs.

Initial Sheet & Drive Setup

python -m commands.cli setup

This creates the column headers in your Google Sheet and the Drive subfolder structure (Research, Scripts, Thumbnails, Assets, Published) under your GOOGLE_DRIVE_FOLDER.

Registering the MCP Server with Claude Code

Open .mcp.json (project root) and update the paths to match your environment:

{
  "mcpServers": {
    "youtube-content": {
      "command": "/absolute/path/to/.venv/bin/python",
      "args": ["-m", "mcpserver.server"],
      "cwd": "/absolute/path/to/youtube-mcp",
      "env": {
        "PYTHONPATH": "/absolute/path/to/youtube-mcp"
      }
    }
  }
}

Claude Code auto-discovers .mcp.json when you open the project directory. Alternatively register it manually:

claude mcp add youtube-content /path/to/.venv/bin/python -- -m mcpserver.server

Restart Claude Code and confirm the server appears under /mcp.

Using the System

Once the MCP server is connected, talk to Claude Code naturally:

What you say	What happens
"Research 5 new ideas about US Taxes"	Claude checks for duplicates, gathers web/news/Reddit data, synthesizes a research doc, and saves it to Drive
"Write a script for row 14"	Claude loads the research doc and script prompt, writes the script, saves it to Drive, and updates the sheet
"Generate visuals for row 14"	Claude loads the script, writes scene-by-scene visual directions, saves to Drive
"What needs a script?"	Returns all videos at Research Complete status
"What are we publishing next week?"	Reads the sheet and shows the publishing calendar
"Add this topic: Property Tax Appeals"	Checks for duplicates, then adds the row

CLI Commands

You can also drive the system from the terminal without Claude Code:

content setup                              # initialize sheet and Drive
content list                               # list all videos
content list --status "Research Complete"  # filter by status
content schedule                           # show publishing calendar
content pending                            # pipeline snapshot
content stats                              # counts by status
content research --topic "Property Tax"    # build a research brief
content script --row 14                    # load script context
content visuals --row 14                   # load visual context

Customizing Prompts

All AI prompt templates are plain Markdown files in prompts/. Edit them to change tone, structure, or guidelines without touching any code:

File	Purpose
research.md	Research document structure
script.md	Script framework and writing style
visuals.md	Scene-by-scene visual direction format
thumbnail.md	Thumbnail concept generation
title.md	Title generation with SEO guidance
description.md	YouTube description writing

Google Sheet Schema

Sheet name: Videos

Column	Description
ID	Auto-assigned row ID
Topic	Video title / topic
Category	Content category
Status	Workflow state (see below)
Priority	High / Medium / Low
Research Score	0–100 quality score
Research Doc	Drive link to research document
Script Doc	Drive link to script document
Thumbnail	Thumbnail concept notes
Publish Date	Scheduled publish date
Created At / Updated At	Timestamps
Video URL	Live YouTube URL after publishing
Notes	Freeform notes

Status flow: Idea → Researching → Research Complete → Script Writing → Script Complete → Thumbnail Ready → Scheduled → Published → Archived

Project Structure

youtube-mcp/
├── mcpserver/        # MCP server (tools Claude Code calls)
├── services/         # Google API clients (Sheets, Docs, Drive, Auth)
├── agents/           # Agent logic (research, script, visual, analytics)
├── commands/         # CLI entry points
├── config/           # Settings and cached OAuth token
├── prompts/          # Editable AI prompt templates
├── logs/             # Rotating daily log files
├── cache/            # Research cache (TTL configurable in .env)
├── .mcp.json         # MCP server config for Claude Code
├── .env.example      # Environment variable template
└── requirements.txt  # Python dependencies

Troubleshooting

MCP server not connecting — check that the command path in .mcp.json points to the Python binary inside your .venv, and that cwd is the absolute project path.

Google auth errors — delete config/token.json and re-run python -m commands.cli auth client_secret.json.

"No valid Google credentials" — make sure GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, and GOOGLE_REFRESH_TOKEN are all set in .env.

Web search returning no results — the system falls back to DuckDuckGo when BRAVE_SEARCH_API_KEY is unset, but Brave gives better results.