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.
README
Hooklayer MCP
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, awould_fail_becausecounterfactual, and aqualityhealth field.predict_viralityruns an independent adversarial check.analyze_account.recommended_chainsteps now exposeconfidence,cost,action_class(authority taxonomy), andexpected_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 SDKexamples/python-example.py— Python usage viaanthropic-mcpclientexamples/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.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器