Turn any UI animation into Framer Motion code with Claude Code

You've seen it happen: you show your AI agent a beautiful animation and ask for “something like this,” and it produces a sad 300ms linear fade. Not because the model can't write animation code — because describing motion in words loses exactly the information that makes motion good: the timings, the easing, the order, the overlap.

MotionLens fixes the input. It watches the animation and hands your agent a Motion Blueprint — a structured spec of every animated element with real values — over MCP, so the agent reads it natively.

1. Sign in at motionlens.dev (5 free credits, no card) → open MCP in the sidebar → create an API key.

2. Add the server to Claude Code:

claude mcp add --transport http motionlens https://motionlens.dev/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Cursor and other MCP clients take the same URL + header in mcp.json. Nothing to install or run locally — it's one hosted endpoint.

Point it at a live site:

> analyze the first 10 seconds of https://feather.computer with MotionLens
  in precision mode, then recreate the hero animation in Framer Motion

Or at any clip URL — a screen recording, a Dribbble shot's video, a Loom export:

> analyze https://cdn.example.com/checkout-flow.mp4 with MotionLens and
  implement the card entrance in our design system

The agent gets back the blueprint: a summary, the beat-by-beat choreography, per-element motions (what moved, from → to, duration, easing, what fired together), feel tags, and an agent-ready prompt. Then it writes code from real values instead of vibes. Every run also lands in your dashboard History with a replay and an annotated storyboard.

• analyze_url — records a live page headlessly (real frame timing) and analyzes it. Auto-playing motion only: load choreography, marquees, loaders. No cursor, so hover/click flows won't appear.
• analyze_clip — direct video URL (mp4/mov/webm). Use this for interactions you recorded yourself.
• get_analysis / list_analyses — free reads over your saved blueprints, so your agent can reference past analyses anytime.

Analyses read video sampled at 8–12fps: choreography, curves, and timing come out accurately; exact spring constants and single-frame micro-details don't — and the blueprint says so in its blind spots section rather than making numbers up. Standard mode costs 2 credits (blueprint + storyboard, clips ≤30s); Precision costs 3 (multi-pass verification, clips ≤10s). Failed runs refund automatically.