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.
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.
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”
0:00
0:04
0:11
0:19
0:28
0:37
Video structure · functional segments
Content taxonomy · 33 dimensions · 8 groups
{
"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_analysisPer-video analysis
300+ creative variables read from every video
get_brand_blueprintBlueprints per brand
Positioning, hooks and patterns, compiled per account
get_top_postsBoost-checked rankings
Every breakout checked against the Meta Ad Library
search_contentSemantic 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.
Mint a key
Settings → Integrations → Claude & Cursor (MCP) → New key. It starts hmk_ and is shown once.
Paste it in
Claude: Settings → Connectors → Add custom connector. Cursor: .cursor/mcp.json. The server URL sits on the same card.
Let your agents call
Workspace-scoped tools. The one write, adding accounts, asks for confirmation before it bills. Revoke any key in one click.
{
"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.
Keep exploring
The rest of the workspace.
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.