AI Video Pipeline MCP Server

An MCP server that powers an AI video production pipeline for algorithm-explanation YouTube videos, designed to run alongside Claude Desktop.

How It Works

Claude Desktop is the AI. This server provides structured tools and prompt templates — no external AI API calls, no API keys required.

When you call a tool (e.g. generate_script), it:

1. Validates your inputs
2. Assembles a rich, domain-specific prompt using carefully crafted templates
3. Returns a prompt packet to Claude Desktop

Claude reads the packet and generates the output (script, scene list, image prompt, Python code) directly in the conversation.

Pipeline

1. generate_script        →  Full narration script with visual cues
2. [record voiceover]     →  Human step — export timestamped transcript
3. plan_scenes            →  Visual scene list with timestamps
4. generate_image_prompt  →  Whisk/Midjourney prompt per image scene
5. generate_video_compositor → Python script that assembles the final video

Tools

Tool	Description
generate_script	Write a complete audio-first narration script
plan_scenes	Convert a timestamped transcript into a scene storyboard
generate_image_prompt	Generate a Whisk/AI image prompt for a scene
generate_video_compositor	Generate a Python video compositor script

Installation & Claude Desktop Setup

Prerequisites

• Python 3.12+
• uv package manager
• Claude Desktop

Install

# Clone the repo
git clone https://github.com/YOUR_USERNAME/mcp_deployment.git
cd mcp_deployment

# Install dependencies
uv sync

Configure Claude Desktop

Add this to your claude_desktop_config.json (usually at %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "ai-video-pipeline": {
      "command": "uv",
      "args": [
        "--directory",
        "C:/Proj/DD/mcp_deployment",
        "run",
        "ai-video-mcp"
      ]
    }
  }
}

Note: Change the --directory path to wherever you cloned the repo.

Restart Claude Desktop. The 4 tools will appear in the tool list.

Testing Without Claude Desktop

Use the MCP Inspector to test tools interactively:

mcp dev src/mcpserver/server.py

Open http://localhost:5173, select a tool, fill in the inputs, and click Run.

---

Re-publishing (GitHub)

To update the server after making changes:

1. Bump the version

In pyproject.toml, increment version:

version = "0.2.1"  # or 0.3.0 for bigger changes

2. Build

uv build

This produces dist/ai_video_pipeline_mcp-X.Y.Z-py3-none-any.whl.

3. Commit and push

git add -A
git commit -m "chore: bump to vX.Y.Z — describe your changes"
git push origin main

4. Update Claude Desktop config (if path changed)

Restart Claude Desktop to pick up the new version. Since you're running from source via uv run, this happens automatically — no reinstall needed.

Optional: Install from GitHub directly

If you want to install the server as a package from GitHub (instead of running from source):

uv pip install git+https://github.com/YOUR_USERNAME/mcp_deployment.git

Then use ai-video-mcp as the command in claude_desktop_config.json (no --directory needed).

---

Project Structure

src/mcpserver/
├── server.py                   # FastMCP server — registers all tools
├── __main__.py                 # Entry point (ai-video-mcp command)
├── tools/
│   ├── script_generator.py     # Tool 1: generate_script
│   ├── scene_planner.py        # Tool 2: plan_scenes
│   ├── image_prompt_generator.py  # Tool 3: generate_image_prompt
│   └── animation_generator.py  # Tool 4: generate_video_compositor
└── utils/
    └── schemas.py              # Pydantic input schemas