youtube-script-mcp

A remote MCP server exposing 4 tools for YouTube script/title strategy, connectable from Claude, ChatGPT, or any MCP client.

Tools

1. analyze_competitor_titles — channelUrl, optional keyword. Pulls recent videos, finds outlier (above-average) titles, returns winning patterns, oversaturated patterns, and fresh title recommendations.

2. generate_channel_dna — bio, optional competitorUrls (up to 3). Returns 1-4 pillars, 1-4 title formulas, a positioning gap, and a channel DNA summary.

3. generate_hooks — niche, optional competitorUrls (up to 3), optional count (1-4). Returns original hook types inspired by competitor patterns (never copies wording).

4. generate_script — scriptStructure, pillar, formula, hookType, titleIdea. Returns a full script draft.

Related MCP server: MCP YouTube Intelligence

Setup

npm install
cp .env.example .env
# fill in .env with your keys
npm run build
npm start

For local dev without building: npm run dev

Required API keys

• YOUTUBE_API_KEY — YouTube Data API v3 key from Google Cloud Console. Enable "YouTube Data API v3" on the project, then create an API key.

• OPENAI_API_KEY — from platform.openai.com.

• SERPER_API_KEY — from serper.dev. Optional but recommended: without it, saturation checks in analyze_competitor_titles are skipped.

Deploying on Replit

1. Add the three keys above as Replit Secrets (not hardcoded).

2. Set the run command to npm run build && npm start.

3. Once running, your MCP endpoint is https://<your-repl-domain>/mcp.

Connecting from Claude / ChatGPT

Both accept a remote MCP server URL. Give them:

https://<your-deployed-domain>/mcp

as the connector URL. The server uses the Streamable HTTP transport (stateful sessions via the Mcp-Session-Id header), which both Claude and ChatGPT's remote-connector support expects.

Notes

• resolveChannelId in src/youtube.ts accepts @handle URLs, /channel/UC... URLs, legacy /c/ and /user/ URLs, or a raw channel ID.

• All AI-generated fields (patterns, pillars, hooks, scripts) come from OpenAI's gpt-4o-mini — swap the model string in src/openai-client.ts if you want a different one.

• This has not been run against live API keys in this environment (no network access here) — test locally with real keys before relying on it in production.