Hook

Solutions · AI companies

Creative intelligence your agents can call.

Hook's analyzed-video corpus and brand graph, exposed as MCP tools inside Claude and Cursor. Your agents ship content decisions grounded in real category data.

Free plan, no credit cardReal virality, not boostedSet up in 2 minutes
Claude Hook · MCP
Why is Halftide out-reaching Northbound on organic?
Hookget_top_posts

Across your set, POV demo reels travel furthest. Northbound sits #2 of 7 on reach-per-follower. You punch above your size. The real gap is paid: 1 live ad vs ~6 per rival.

Reply to Claude…Opus 4.7

One answer over MCP · workspace-scoped tools

Build vs buy

You started building this.

Teams shipping content automation on Claude and Cursor keep arriving at the same missing layer. The instinct is right. Finishing it is the expensive part.

the signal

The need is validated

If your team ever sketched a competitor-content scraper, you already proved the demand. The only question left is build vs buy, and which gets your agents grounded first.

the gap

Your agents stop at text

Claude drafts the script, but it has never watched a reel. Hooks, pacing, cuts, captions: the variables that decide reach live in the video layer your stack can't see.

the moat

Scraping is the easy 10%

Pulling posts is a weekend project. The taxonomy is not: 300+ creative variables per video, boost-checks against the Meta Ad Library, blueprints that update as the category moves.

The corpus

Every video in your category, already analyzed.

Hook watches Instagram, TikTok and the Meta Ad Library, reads 300+ creative variables from every video, and compiles a creative blueprint per brand. Scores are boost-checked, so a bought spike never reads as a trend.

@halftide0:4214 shots detected212K views“You have been storing it wrong”

Hook

0:00

Context

0:04

Demo

0:11

Demo

0:19

Proof

0:28

CTA

0:37

Video structure · functional segments

Hook0:00-0:03Direct claim over product close-up
Context0:03-0:11Problem framing with voiceover
Demo0:11-0:28Hands-on walkthrough, three cuts
Proof0:28-0:37Before and after, side by side
CTA0:37-0:42Verbal ask plus pinned caption

Content taxonomy · 33 dimensions · 8 groups

video-analysis.json
{
  "video_structure": ["Hook 0:00", "Context 0:03", "Demo 0:11", ...],
  "tags_json": {
    "hook_type": "direct_claim",
    "visual_style": "product_closeup",
    "audio": "voiceover_trending",
    "cta_type": "verbal_plus_caption",
  },
  "shots": 14
}

Every reel in the niche gets this treatment · frame by frame

get_post_analysis

Per-video analysis

300+ creative variables read from every video

get_brand_blueprint

Blueprints per brand

Positioning, hooks and patterns, compiled per account

get_top_posts

Boost-checked rankings

Every breakout checked against the Meta Ad Library

search_content

Semantic search

Query the whole corpus in plain language

Connect

Build on it. Don't rebuild it.

Mint a workspace key, paste it into Claude or Cursor, and your agents get Hook's workspace-scoped tools over MCP. Reads are just reads. The one write is confirmation-gated, and keys revoke in one click.

01

Mint a key

Settings → Integrations → Claude & Cursor (MCP) → New key. It starts hmk_ and is shown once.

02

Paste it in

Claude: Settings → Connectors → Add custom connector. Cursor: .cursor/mcp.json. The server URL sits on the same card.

03

Let your agents call

Workspace-scoped tools. The one write, adding accounts, asks for confirmation before it bills. Revoke any key in one click.

Cursor · .cursor/mcp.json
{
  "mcpServers": {
    "hook": {
      "url": "<server URL from your MCP card>",
      "headers": { "Authorization": "Bearer hmk_YOUR_KEY" }
    }
  }
}

Works with Claude · Claude Code · Cursor · any MCP client

Proof

The same depth, run on AI-native brands.

Public teardowns of the organic engines at ElevenLabs and Replit. Analyses, not endorsements: every number renders straight from the corpus.

FAQ

Questions, answered.

You've validated the need. The question is which gets you there faster. Scraping is the easy 10%; the analysis taxonomy, the boost-checks against the Meta Ad Library and the blueprint aggregation are years of iteration. Connect the MCP server and your agents get all of it this afternoon.

A creative-intelligence coworker in your Slack

Stop pasting TikTok links into ChatGPT.

You're doing the watching for a model that can't see video. Hook already watched your category, checked what was boosted, and drafts the next one in your Slack.