FrameFetch

Any social video → data, in one call.

Give a video URL, get metadata, transcript, insights, and parametrically-sampled frames. Built API-first and MCP-first for AI agents — pay per call.

YouTube · Shorts · TikTok · Instagram · Pinterest · Reddit

What you get

metadata (title, author, duration, date) insights (views, likes, comments) transcript (captions or Whisper) frames — every / every-Nth / 1-per-sec / range, any size

Quickstart (REST)

curl -X POST https://framefetch.net/v1/extract \
  -H "Authorization: Bearer <your-key>" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://www.youtube.com/watch?v=...",
    "fields": ["metadata", "transcript", "frames"],
    "frames": { "mode": "fps", "fps": 1, "width": 480 }
  }'

Returns one JSON object: metadata, insights, transcript, and frames[] as time-limited signed image URLs, plus a cost block. Request only the fields you need.

Endpoints

Method · PathPurpose
POST /v1/extractExtract data from a video URL
GET /v1/platformsCapability matrix per platform
GET /healthzHealth check
POST /mcpMCP (Streamable HTTP) for agents

For AI agents (MCP)

FrameFetch ships an MCP server with two tools: framefetch_extract and framefetch_platform_capabilities. Add it to your agent and call a video URL directly. Machine docs: /llms.txt.

Pricing

Pay per call — light calls (metadata/captions) are sub-cent; transcript is metered per minute; frames are metered per frame. A free tier covers browsing/light use. Pay with prepaid credits or x402 (USDC) with no account.

Why FrameFetch

The only API that returns parametric frames from a social-video URL, includes Reddit-video transcripts, and lets an agent pay without signing up. One clean schema across all six platforms.