Refs

Product Introduction

1. Definition: Refs is a structured video reference library and Model Context Protocol (MCP) server specifically designed for AI agents like Claude. It is a technical database of over 3,000 successful marketing and product films (launch videos, explainers, motion pieces) that have been deconstructed frame-by-frame into quantifiable blueprints.
2. Core Value Proposition: It exists to solve the "garbage in, garbage out" problem for AI-generated video content. By providing Claude with precise, data-driven references—including shot-by-shot timing, editing formulas, and production recipes—it enables the AI to plan and generate video concepts based on proven, high-performing cinematic patterns rather than vague prompts.

Main Features

1. Frame-Accurate Video Deconstruction: Every film in the archive is analyzed using computer vision and audio processing to detect exact shot boundaries (cuts). This results in a precise "cut map" displaying metrics like cuts per minute (CPM) and average shot length (ASL), providing a quantitative blueprint of the film's pacing and structure.
2. 15-Plate Analytical Blueprint: Each video is summarized into a standardized 15-part analysis. This includes the hook (first two seconds), format classification, color palette, sound design, script breakdown, and a timed beat-by-beat outline of the entire narrative and visual flow.
3. Runnable Prompt Recipes & MCP Integration: The core technical feature is the generation of executable "recipe" prompts. These are detailed, step-by-step instructions (e.g., "Use Remotion to create an original 15-second, 60 fps motion-design showreel...") that can be fed directly into Claude Code or Cursor. The Refs MCP server allows Claude to search, retrieve, and utilize these blueprints and recipes autonomously within the developer's IDE.

Problems Solved

1. Pain Point: The creative bottleneck and quality inconsistency in producing marketing videos. Teams struggle to articulate or deconstruct what makes a reference video effective, leading to vague briefs and AI outputs that miss the mark on pacing, emotional impact, and structural integrity.
2. Target Audience: Product marketers, growth teams, indie founders, and developer-entrepreneurs who need to produce high-conversion launch films or explainers but lack dedicated motion design resources or cinematic expertise.
3. Use Cases: A founder using Claude Code to generate a launch video script and Remotion code based on the "Founder-Led Launch Film" formula. A marketing team analyzing the "Dark Staged Product Reveal" pattern to brief an agency. A developer prompting an AI agent to create a product demo video with the exact pacing and shot structure of top-performing SaaS ads.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Transforms subjective "vibes" into actionable, data-driven instructions for AI. Vast, curated library of proven references from top tech companies (Apple, Google, OpenAI, etc.). Deep integration with Claude's development workflow via MCP enables a closed-loop creation process.
   • Cons: The output quality is still dependent on the user's ability to effectively utilize the AI agent (Claude) and the underlying technical stack (e.g., Remotion, video editing software). It provides the blueprint and recipe, not the final rendered asset. The library is focused on modern tech/product marketing films, not other video genres.

2. Key Alternatives & Differentiation:

   • Traditional Mood Boards (Pinterest, Are.na): These are purely visual and inspirational, lacking quantitative data, timed beats, or executable prompts. Refs adds a layer of analytical depth and AI-actionable structure.
   • Video Stock Libraries (Artgrid, Storyblocks): Provide raw footage, not deconstructed analyses of finished, successful narratives. They are a component supplier, while Refs is a strategic blueprint provider.
   • Creative Agencies: Offer full-service production at high cost and slower speed. Refs empowers in-house teams and individuals to achieve agency-level strategic thinking at a fraction of the cost and time, though it requires their own execution.

Frequently Asked Questions (FAQ)

1. How does Refs work with Claude AI? Refs operates as an MCP (Model Context Protocol) server that connects directly to Claude Code, Claude Codex, or the Cursor IDE. Once connected via a command-line install, Claude can search the entire Refs database, retrieve detailed film blueprints and prompt recipes, and use that structured reference material to plan and help build your custom video project.
2. What is the difference between the free and Pro plan? The free plan allows you to browse the entire archive, view videos with cut maps and basic metrics, and see the first few plates of each blueprint. The Pro plan unlocks all 15 analytical plates for every film, provides the complete runnable prompt recipes, grants access to comparative "formulas," and includes the MCP integration with a monthly credit allowance for queries.
3. Does Refs host or own the video content? No. Refs does not host any video files. All films are embedded from and link directly to their original source (like YouTube or a company's site). Refs provides its own layer of frame-accurate measurement, analytical metadata, and commentary, acting as a critical analysis and reference tool rather than a content repository.
4. What kind of videos are in the Refs library? The library is curated around high-performance product marketing and launch films from leading technology and design-driven companies. This includes product launch teasers, capability demos, brand films, explainer videos, and motion design showreels from companies like Apple, Google, Anthropic, Linear, and Vercel.
5. Can I use Refs without being a developer? Yes, the archive is freely browseable for inspiration and analysis by marketers, designers, and product managers. However, to fully leverage the AI-integrated "build" functionality—using the prompt recipes with Claude—technical familiarity with AI coding assistants and development environments is necessary.