n8n AI Agent 集成

You are an AI research agent integrated into an n8n workflow with access to Anysite MCP tools.
Use these tools to extract data from LinkedIn, Instagram, Twitter, Reddit, YouTube, SEC, and Y Combinator.

When the user requests information:
1. Identify the appropriate MCP tool(s) to use
2. Extract the data using the tools
3. Analyze and structure the information
4. Return JSON-formatted results for the workflow

Available platforms: LinkedIn, Instagram, Twitter, Reddit, YouTube, SEC EDGAR, Y Combinator, Web Parser

Always provide accurate, well-structured data that can be processed by subsequent workflow nodes.

Webhook → AI Agent (MCP) → Enrich Contact → Save to CRM

"Research {{$json.email}} on LinkedIn. Find their profile, current company,
job title, and recent activity. Return structured JSON."

Schedule (Daily) → AI Agent (MCP) → Format Report → Send to Slack

"Monitor competitor 'TechCorp' on LinkedIn:
- Get last 7 days of posts
- Extract engagement metrics
- Identify trending topics
Return JSON with summary."

Manual Trigger → AI Agent (MCP) → Split Data → Parallel Processing → Merge Results → Output

"Find all mentions of 'AI automation tools' across:
- LinkedIn (posts from last week)
- Twitter (tweets from last week)
- Reddit (discussions from last month)
- YouTube (videos from last 3 months)

Analyze sentiment and provide summary."

HTTP Request → AI Agent (MCP) → Extract Financial Data → Store in Database

"Get latest 10-K and 10-Q filings for {{$json.company_name}}.
Extract key financial metrics, risk factors, and MD&A summary."

Airtable Trigger → AI Agent (MCP) → YC Research → Update Airtable

"Research YC company {{$json.slug}}:
- Get company profile
- Find founders and backgrounds
- Identify batch and status
Return structured data for database."

// 获取前一个 AI Agent 节点的数据
{{ $json.research_results }}

// 访问特定字段
{{ $json.linkedin_profile.name }}
{{ $json.company_data.employee_count }}

// 在 AI Agent 系统提示词中
"If the request is about LinkedIn, use linkedin_* tools.
If it's about SEC filings, use search_sec_companies and get_sec_document.
If it's about YouTube, use youtube_* tools.
Always choose the most appropriate tool for the data source."

"Return your findings as JSON with this structure:
{
  'summary': 'Brief overview',
  'data': { extracted data },
  'sources': ['tool1', 'tool2'],
  'timestamp': 'ISO format'
}
"

"Monitor these 3 competitors on LinkedIn:
- Company A
- Company B
- Company C

For each, get:
- Last 5 posts
- Engagement metrics
- Employee growth trends

Return JSON report."

"Research {{$json.topic}} across all platforms:
- LinkedIn posts (last week)
- Twitter mentions (last week)
- Reddit discussions (last month)
- YouTube videos (last 3 months)

Analyze sentiment and trends."

Trigger
  → Agent 1: LinkedIn Research (MCP)
  → Agent 2: Twitter Sentiment (MCP)
  → Agent 3: Synthesis & Analysis
  → Send Report

Input Data
  → IF node (has LinkedIn URL?)
      → Yes: AI Agent extracts LinkedIn data
      → No: AI Agent searches by email/name
  → Merge results
  → Continue workflow

Get Task
  → AI Agent (MCP): Initial Research
  → Store in n8n Database
  → IF: Need more data?
      → Yes: AI Agent (MCP): Deep Dive
      → Loop back
  → Final Output

Schedule (Every hour)
  → AI Agent: Check for new posts/filings
  → IF: Found new content?
      → Process and alert
      → Update tracking database
  → Sleep until next run