Custom MCP for Lenny's Podcast
Building a Product Intelligence Server for elite PM frameworks
Deliverables / Skills Utilized
🔍 The Problem
PM frameworks are trapped in transcripts. The best product advice from interviews with top PMs—pricing strategies, growth tactics, roadmap frameworks—sits buried in hours of podcast content.
Context-switching between IDE and browser to find relevant advice slows down technical drafting. Every time I'm writing a PRD or strategy doc, I'd think "Lenny had a great episode on this..." and lose 20 minutes searching.
💡 The Solution
A custom Model Context Protocol (MCP) server that indexes 320+ Lenny's Podcast transcripts, allowing an AI agent to search for elite PM advice in real-time—directly from my IDE.
🛠️ Technical Implementation
📥 Data Acquisition
- Cloned 320+ markdown transcripts from ChatPRD's public archive
- Organized by guest, topic, and date for efficient indexing
- 2M+ words of PM wisdom, locally accessible
🖥️ MCP Server Development
- Built Python-based server using the official MCP SDK
- Implemented semantic search across all transcripts
- Handles concurrent queries with low latency
🔧 Custom Tool: search_lenny_insights
- Query interface: natural language questions about PM topics
- Returns relevant excerpts with episode context
- Supports filtering by guest, topic, or date range
🤖 Claude Code Integration
- Connected via --mcp flag for agentic workflows
- Real-time access during PRD drafting
- No browser context-switching required
💬 Example Usage
Query: "What does Shreyas Doshi say about high-agency PMs?"
Response: Found 3 relevant excerpts from Shreyas Doshi episodes discussing how high-agency PMs don't wait for permission—they identify the most important problem and start solving it.
📊 Impact
| Metric | Result |
|---|---|
| 📚 Transcripts Indexed | 320+ |
| 📝 Total Content | 2M+ words |
| ⚡ Search Latency | Under 2s |
| 🎯 Context Switches Saved | 100% |
🎯 Why This Matters
Transformed a static archive into an active tool for:
- 📄 Writing PRDs with real framework references
- 🧠 Brainstorming technical strategy with expert backing
- 📈 Learning PM craft through targeted retrieval
- ⚡ Building "just-in-time" knowledge systems
🌐 The Bigger Picture
This project demonstrates the power of MCP for building personal productivity tools. Any knowledge corpus—books, courses, internal docs—can become an AI-accessible resource with the same pattern.
🔧 Stack
- Python: Core server implementation
- MCP SDK: Model Context Protocol framework
- Claude Code: AI agent integration
- Markdown: Transcript storage format