ai-study

2026-05-11-sf67-content-creation-mcps

7 min read

5 Content Creation MCPs for Claude — Synthesis

Source: AI Basic Series, “SF67 — 5 Content Creation MCPs for Claude: Install Links + Starter Prompts” (PDF, no date on the document; ingested 2026-05-11). Authors: Not named on the document. Branded “BUILT WITH CLAUDE CODE | AI-BASIC-SERIES.VERCEL.APP”. Underlying material: Promotional one-pager for the paid “Actionable AI Accelerator” course at whop.com/actionable-ai/standard-3b. Practical content is a 5-item setup guide for MCP servers.

Caveat on retrieval: The PDF has no explicit publication date or author byline. The series tag “SF67” suggests a numbered series. The document is overtly promotional (two CTAs pointing at the paid course). The technical claims below have been preserved verbatim from the PDF; install commands and namespaces have not been independently verified against the upstream packages and should be checked before use.

Headline message

A short, branded setup guide listing five MCP (Model Context Protocol) servers a content-creation workflow could install into Claude, with a one-line install snippet and a starter prompt for each. The framing pitch — useful in itself — is that the value of MCPs is operational: Claude doesn’t just describe a tool, it logs into it and uses it. The document closes by emphasising that the real leverage is chaining MCPs across a single Claude conversation rather than treating each as standalone.

Key takeouts

  • MCPs reframed as “abilities.” The opening definition: MCPs are plugins that give Claude access to external tools and services; “It’s not just reading about it, it’s logging in and using it.” Useful framing when explaining MCPs to a non-technical audience.
  • The five servers selected for content workflows: NotebookLM, Remotion, SupaData, Stitch, and Magic (21st.dev) — covering knowledge bases, programmatic video, social-video transcripts, UI mockups, and React/Tailwind component generation respectively.
  • Each entry includes a starter prompt, not just install. This is the practical asset of the guide: a tested phrasing that gets useful output from each MCP on first run.
  • The pitch is chaining. Three example chains are given (e.g. SupaData → NotebookLM → Remotion; Stitch → Magic; multi-transcript SupaData → NotebookLM infographic). These illustrate the multi-MCP-per-conversation pattern.
  • Commercial context. The PDF exists to promote a paid course (“55+ AI skills. 10+ automated agents. One-command installs.”) at whop.com/actionable-ai/standard-3b. The MCP list itself is a free-tier lead magnet.

Wider context

This sits in the wider Claude MCP ecosystem. MCPs (Model Context Protocol) are configured per-Claude via ~/.claude.json mcpServers entries, an npx or npm command, and (often) an env-var holding an API key. The five tools covered here are representative of a common workflow archetype — content marketer / solo creator / agency-of-one — and demonstrate the typical install ergonomics (single shell command + a JSON block in a config file).

The doc does not address: security review of third-party MCPs, what data each MCP sends where, vendor lock-in or paid-tier costs (Supadata, Stitch, Magic all reference an API_KEY env var; pricing is not in the PDF), or any comparison with non-MCP equivalents.

Section-by-section breakdown

1. What MCPs are (opening definition)

“MCPs (Model Context Protocol) are plugins that give Claude access to external tools and services. When you connect one, Claude can actually operate the tool on your behalf. It’s not just reading about it, it’s logging in and using it. Think of MCPs as giving Claude new abilities.”

2. NotebookLM MCP

Capability claimed: “Create and manage Google NotebookLM notebooks. Add sources, generate podcasts, infographics, mind maps, slides, and study guides from any content.”

Install (per the PDF):

npm install -g notebooklm-mcp

Config (per the PDF, added to ~/.claude.json mcpServers):

"notebooklm-mcp": {
  "command": "notebooklm-mcp"
}

Starter prompt: “Create a NotebookLM notebook from these 5 blog post URLs. Generate an audio overview and an infographic summarizing the key themes.”

3. Remotion MCP

Capability claimed: “Access the Remotion video framework for creating programmatic videos. Motion graphics, animated text, data visualizations, rendered as real video files.”

Config (per the PDF):

"remotion-documentation": {
  "command": "npx",
  "args": ["-y", "@remotion/mcp@latest"]
}

Starter prompt: “Create a 15-second Remotion animation with a headline fading in, three bullet points appearing one at a time, and a CTA. Dark background, white text, blue accent.”

Note: the JSON key is remotion-documentation in the PDF, suggesting this MCP is scoped to Remotion docs / scaffolding rather than running renders — verify against upstream.

4. SupaData MCP

Capability claimed: “Pull transcripts from YouTube, TikTok, Instagram, X, and Facebook. Also pulls metadata like view counts and likes.”

Setup (per the PDF):

  1. Get API key from supadata.ai
  2. Save key to ~/.config/supadata/.env as SUPADATA_API_KEY=your_key_here
  3. Use via the /transcript skill in Claude Code

Starter prompt: “Pull the transcript from this video [URL]. Break down the hook framework, identify key talking points, and tell me what made this work.”

Author commentary in the PDF: “I use this every day to break down competitor content and steal hooks.”

5. Stitch MCP

Capability claimed: “Generate full UI screens, design systems, and visual mockups from text descriptions. Describe a page and Stitch builds it with proper layout and visual hierarchy.”

Config (per the PDF):

"stitch": {
  "command": "npx",
  "args": ["@_davideast/stitch-mcp", "proxy"],
  "env": { "STITCH_API_KEY": "your_key" }
}

Starter prompt: “Create a landing page screen for a free PDF download. Include headline, 4 value props, email capture form, and PDF preview. Modern dark theme, blue accent.”

6. Magic (21st.dev) MCP

Capability claimed: “Generate production-ready React + Tailwind components from descriptions. Buttons, forms, dashboards, navigation bars, pricing tables, entire page sections.”

Config (per the PDF):

"magic": {
  "command": "npx",
  "args": ["-y", "@21st-dev/magic@latest"],
  "env": { "API_KEY": "your_21st_api_key" }
}

Starter prompt: “Generate a pricing table component with 3 tiers, feature comparison checkmarks, and a highlighted Most Popular badge. Clean modern style, rounded corners.”

7. Chaining MCPs together

Three composition examples given by the author:

  1. Transcript → KB → Video. SupaData pulls a transcript; NotebookLM builds a knowledge base from it; Remotion generates a video summary.
  2. Layout → Code. Stitch produces the landing page layout; Magic builds the actual components.
  3. Competitive analysis pipeline. Pull 5 competitor video transcripts with SupaData; create a NotebookLM notebook from all of them; generate an infographic comparing strategies.

Author’s framing: “Each chain runs in one Claude conversation. No switching between tools.”

Action implications / open questions

For someone evaluating this as a workflow recipe:

  • Verify each package and namespace before installing. PDF snippets are presentational and may lag the upstream package name, scope, or required args. In particular @_davideast/stitch-mcp and the remotion-documentation key both warrant a check against current docs.
  • Audit data egress per MCP. SupaData, Stitch and Magic all require an API key — that means content/prompts leave the local machine. For sensitive material (client work, draft content) understand each vendor’s data handling.
  • Cost model is not in the PDF for any of the paid services. Establish that before chaining them, especially in agent loops.
  • The promotional framing is part of the document. The free guide is a lead magnet for a paid course — that doesn’t make the technical pointers wrong, but it does mean the curation is optimised for “wow factor in one Claude session” rather than for long-term durability or open-source-only stacks.

Open questions:

  • Where does this sit relative to first-party Anthropic-published MCPs? The PDF doesn’t compare.
  • Are the chains genuinely robust in one Claude session, or do they require explicit context management (e.g., persisted artefacts in a project directory)?
  • Does the publication have a verified date and author somewhere outside the PDF (newsletter archive, X post)? The PDF itself carries neither.

Graph View

Backlinks