hooklayer

hooklayer

Viral-content intelligence for AI agents — 7 read-only MCP tools for TikTok/YouTube hook scoring, virality prediction, trend analysis, and viral template search, with evidence-layer scoring.

Category
Visit Server

README

Hooklayer MCP

License: MIT npm version Version MCP Production Tools OAuth smithery badge

Viral-content intelligence for AI agents. Drop the Hooklayer MCP server into Claude Desktop, Cursor, n8n, or any HTTP MCP client and your agent gets 7 read-only tools for scoring hooks, remixing viral videos, surfacing live trends, and analyzing TikTok creators. The flagship analyze_account tool includes a recommended_chain field with suggested follow-up calls the agent can review and execute.

7 read-only tools · all return structured JSON · no mutations · no side effects

v1.1.0 (2026-05-14): The evidence layer ships. Every score includes signals[] with cited evidence, a would_fail_because counterfactual, and a quality health field. predict_virality runs an independent adversarial check. analyze_account.recommended_chain steps now expose confidence, cost, action_class (authority taxonomy), and expected_output. See CHANGELOG.md for the full ship.


🚀 Quick install

Claude Desktop

Claude Desktop doesn't natively support remote HTTP MCP servers — it needs the mcp-remote bridge. Two ways to install:

Option 1 — Custom Connector (easiest, no config file edit)

In Claude's web/desktop UI: Settings → Connectors → Add custom connector → paste this URL:

https://hooklayer.dev/api/mcp

Claude.ai will walk you through OAuth (no manual key paste). Done.

Option 2 — Direct config (for power users who want hl_live_ key auth)

Edit claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "hooklayer": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@0.1.38",
        "https://hooklayer.dev/api/mcp",
        "--header",
        "Authorization:Bearer hl_live_..."
      ]
    }
  }
}

Get your free hl_live_ key at https://hooklayer.dev/auth/signup — 100 lifetime credits, no card required.

Restart Claude Desktop. The 7 Hooklayer tools appear in the 🔌 connector list.

Cursor

~/.cursor/mcp.json:

{
  "mcpServers": {
    "hooklayer": {
      "url": "https://hooklayer.dev/api/mcp",
      "transport": "http",
      "headers": {
        "Authorization": "Bearer hl_live_..."
      }
    }
  }
}

n8n

In your workflow, add an MCP Client node and configure as a remote HTTP MCP server:

  • URL: https://hooklayer.dev/api/mcp
  • Transport: HTTP
  • Header: Authorization: Bearer hl_live_...

All 7 tools appear in the node's "Tool" dropdown.

OAuth 2.1 + PKCE (for Claude.ai connector + custom apps)

Hooklayer is fully OAuth 2.1 compliant — discovery, Dynamic Client Registration, PKCE, refresh token rotation. MCP clients that prefer OAuth over API keys work out of the box.

Discovery endpoints (no auth required, machine-readable):

# Authorization server metadata (RFC 8414)
curl https://hooklayer.dev/.well-known/oauth-authorization-server

# Protected resource metadata (RFC 9728)
curl https://hooklayer.dev/.well-known/oauth-protected-resource

Dynamic Client Registration (create a client without a manual signup form):

curl -X POST https://hooklayer.dev/oauth/register \
  -H "Content-Type: application/json" \
  -d '{
    "client_name": "Your MCP client",
    "redirect_uris": ["https://yourapp.com/oauth/callback"]
  }'
# Returns: client_id, client_secret (for confidential clients)

Hitting tools/call without auth returns 401 plus a WWW-Authenticate header pointing at the resource metadata — Claude.ai, Cursor, and other MCP clients use this to auto-discover the OAuth flow.

Other clients

Any HTTP MCP client. Protocol negotiates 2024-11-05 (broadest compat) or 2025-06-18 (Streamable HTTP + structuredContent).

# Quick test — initialize handshake works without auth:
curl -X POST https://hooklayer.dev/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize"}'

🧠 The 7 tools

Tool Credits What it does
analyze_account 5 TikTok creator deep dive: viral DNA scores, format fingerprint, top 5 videos, content gaps, recommended_chain with suggested follow-up tools.
score_hook 1 Score any hook 0-100 against proven viral patterns. Returns 3 rewrites at higher quality.
viral_remix 3 URL or transcript → fresh script with mirrored viral DNA. Scene-by-scene with camera shots.
trend_pulse 1 Real-time rising opportunities + saturated patterns per niche. 12-hour cache.
find_viral_template 1 Niche-fit ranked templates with hook patterns + example URLs.
match_voice 2 Extract a creator's voice DNA from 3+ samples, rewrite a draft in their style.
predict_virality 2 Score a draft script for viral potential before publishing. Retention diagnosis.

Full schemas + curl examples: https://hooklayer.dev/docs


💡 The recommended-chain pattern

analyze_account returns a recommended_chain field with suggested follow-up tools and pre-filled parameters. The agent decides whether and when to execute them:

{
  "viral_dna_score": 87,
  "steal_map": [...],
  "recommended_chain": [
    {
      "tool": "match_voice",
      "params": {
        "draft": "<<<USER_DRAFT>>>",
        "reference_samples": ["https://tiktok.com/...", "...", "..."]
      },
      "reason": "High-signal voice DNA — consistent across top 5 videos"
    },
    {
      "tool": "trend_pulse",
      "params": { "niche": "challenge_videos" },
      "reason": "Verify their formula maps to current trends"
    },
    {
      "tool": "viral_remix",
      "params": { "source_url": "https://tiktok.com/..." },
      "reason": "Their #2 video has the highest copyable structure"
    }
  ]
}

The agent can review these suggestions and call the recommended tools if appropriate. Each tool still requires explicit invocation — nothing runs automatically without agent/user consent.


📦 Examples

  • examples/typescript-example.ts — TypeScript usage via the MCP SDK
  • examples/python-example.py — Python usage via anthropic-mcp client
  • examples/curl-test.sh — Raw curl tests for every endpoint

💵 Pricing

  • Free — 100 lifetime credits at signup, no card
  • Starter — $49/month, 5,000 credits, 60 req/min
  • Pro — $149/month, 25,000 credits, 300 req/min, brand voice memory
  • Agency — $499/month, 150,000 credits, 1,000 req/min, white-label, 10 seats
  • Pay-as-you-go — $25 for 5,000 credits, never expire

Full pricing: https://hooklayer.dev/pricing


🛠 Architecture

Hosted MCP server (no stdio install needed):

Your agent (Claude/Cursor/n8n)
        │
        │  JSON-RPC 2.0 over HTTP
        ▼
https://hooklayer.dev/api/mcp
        │
        ├── initialize / ping / tools/list  (no auth)
        └── tools/call                       (Bearer hl_live_*)
                │
                └── Routes internally to /v1/* REST endpoints
                    100K+ analyzed viral videos
                    ScrapeCreators + Whisper + Sonnet pipeline

Source code for the hosted server lives at hooklayer.dev (closed source — the analysis pipeline is the moat). This repo holds the public client docs, examples, and config snippets.


📚 Links

  • Install page: https://hooklayer.dev/mcp
  • Full docs: https://hooklayer.dev/docs
  • Playground (no signup): https://hooklayer.dev/playground
  • Pricing: https://hooklayer.dev/pricing
  • OAuth metadata: /.well-known/oauth-authorization-server
  • Issues + bug reports: GitHub Issues

🔒 Security

All 7 tools are read-only. They analyze, score, and generate content — none of them create, modify, or delete any user data, accounts, or external resources. Every tool carries readOnlyHint: true and destructiveHint: false in its MCP annotations.

Authentication: tools/call requires a Bearer token (hl_live_* API key or OAuth 2.1 access token). Public methods (initialize, ping, tools/list) work without auth so MCP clients can handshake and discover tools before authentication.

recommended_chain is advisory only. The analyze_account response includes suggested follow-up tools with pre-filled parameters. These are data — the agent and user decide whether to execute them. No tool call ever triggers additional tool calls server-side.

Data handling: Hooklayer processes the inputs you send (handles, hooks, scripts, URLs) to return analysis results. We do not store your content beyond the request lifecycle except for anonymous usage metrics.

To report a security issue: GitHub Issues or email security@hooklayer.dev.


📄 License

MIT — see LICENSE.

The MCP client examples and config snippets in this repo are MIT. The hosted Hooklayer service at hooklayer.dev is a commercial product with the pricing tiers listed above.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured