TL;DR: Last Friday's livestream drew one of our largest audiences in months, and for good reason. Everyone wants to know: What the hell is an MCP server, and how can it accelerate your ability to find pain-based segments at scale? We demonstrated live how to deploy recursive AI agents that generated 1,800 personalized value propositions from the Shovels database in one hour, at seven cents per prospect. This isn't theory, it's your complete guide to our Shovels MCP Server (thank you Jacob Dietel).
Remember when having a data scientist for every prospect was a pipe dream?
Well, wake up. That dream just became reality, and it costs seven cents.
The reason our Friday livestream packed the virtual room wasn't just Doug’s excellent taste in intro music (John Cougar Mellencamp everyone). It's because we demonstrated something that seemed impossible six months ago: deploying recursive AI agents that generated 1,800 pain-based value propositions for construction industry prospects in one hour.
The kicker? The entire system runs on MCP servers, costs pennies per prospect, and keeps getting smarter with each iteration.
The MCP Revolution: When Databases Start Talking Back
Let's start with what changed everything: Model Context Protocol (MCP) servers.
Think of MCP as the bridge between Claude (or any LLM) and your databases. It's not actually a server, it's a connection protocol that lets AI models have natural language conversations with structured data. Instead of writing SQL queries yourself, you ask questions in plain English, and the model figures out how to get the answers.
Here's Jordan's explanation from our livestream:
"It's just a way for a large language model to talk to a database or a tool... The nice thing about recursive intelligence is that the model talks to the tool, talks to the database, the database responds, and the model's like, 'that's not exactly what I wanted. Let me try again. Let me try again. Let me try again.'"
The game-changer: The model eliminates hallucinations because it's not inventing data (gasps) - it's finding it. When Claude talks to the Shovels database containing 170,626,720 building permits, it's not making up project values or contractor names. It's pulling real data and synthesizing real insights.
Live Demo: From 0 to PVP in Four Columns
Our demonstration used the International Builders Conference as the target list—1,800 exhibitors ranging from lumber companies to lawn care services. Each needed different value from the permits database.
The beauty of our system? It only required four columns in Clay:
Column 1: Research the Company
A basic AI agent researches each exhibitor:
What do they do?
How does their business relate to building permits?
What's their relevance score (1-10)?
Example output for 3A Composites USA:
Background: Leading manufacturer of composite panels and materials, flagship brands include DIBOND, GATORFOAM, serving architecture and construction markets.
Shovels Relevance: High permit-relevance score of 8/10. Focus on NEW_CONSTRUCTION and REMODEL permits, particularly high-value commercial projects where aluminum composite materials are specified.Column 2: The MCP Server Prompt
This is where the magic happens. We feed the company research into a prompt that:
Understands the Shovels database structure
Maps business activities to permit types
Defines the recursive search strategy
Column 3: Call Snowflake (The Recursive Agent)
Here's where Jacob's MCP GTM server shines. The agent:
Analyzes the prospect: Identifies core business activity
Selects primary filter: Maps to relevant permit boolean (roofing, HVAC, solar, etc.)
Executes the scalpel query: Runs targeted SQL for top opportunities
Enriches with contractor data: Pulls contact information
Assembles final JSON: Formats actionable leads
The recursion happens automatically. If the first query fails, it pivots. If results aren't good enough, it tries different approaches. The model keeps looping until it gets the right answer.
Column 4: Write the Message
Claude Opus takes all the enriched data and crafts a permissionless value proposition.
The $144M Miami Project: A PVP Breakdown
Let's examine the actual output for 3A Composites USA:
Subject: the $144m Miami project?
The Hook: JOHN M LEETE just pulled a $144M permit for a mixed-use development in Miami FL
The Value: A project this size needs thousands of square feet of facade materials, and your DIBOND panels are perfect for modern mixed-use exteriors.
The Proof: Three additional projects totaling $267.7M in similar developments
The Ask: I have the full contractor list with direct phone numbers. Want me to send you the top 10 opportunities with facade requirements?
This isn't just lead generation—it's intelligence delivery. The prospect would pay to receive this information even if they never bought from Shovels.
Beyond Shovels: Your Data Moat Becomes Your Growth Engine
While we demonstrated with Shovels' construction data, the principles apply to any database:
HubSpot MCP Server: Instead of staring at attribution reports, ask "What should I be worried about this quarter?" or "Which deals are at risk and why?"
Snowflake/Looker Connections: Your data warehouse becomes conversational. No more waiting for data engineers—ask natural language questions and get real-time insights.
Internal Company Data: Connect customer success data, support tickets, and usage analytics. Deploy recursive agents to identify expansion opportunities or churn risks.
The key insight from Jordan: Start with the end in mind.
"The question that I would ask is, if you had the entire team... spend a whole day looking at all of your data for one prospect, what would you do? What would you provide them? What would be incredibly valuable to them?"
The Three GTM Eras: Where We've Been and Where We're Going
Jordan's framework shows how we got here:
Understanding Era (pre-2008)
Direct customer conversations
Deep insights, low scale
Expansion Era (2008-2022)
Cheap capital fuels growth
Omnichannel blitz, retention ignored
The era of "spray and pray"
AI Era (2022-present)
AI co-creates strategy
Budgets scrutinized
Messages adapt in real-time
We're witnessing Jevons Paradox in GTM: As intelligence becomes cheaper and more efficient, we don't use less of it, we use exponentially more. Just like efficient steam engines (Doug knows this) led to more coal consumption, efficient AI leads to more personalized outreach.
The Recursive Intelligence Stack
Here's what you need to build this:
1. MCP Server Setup
Local: Connect directly to your databases via VS Code with Claude
Cloud: Use services like MCPGTM.com for scaled deployment
Cost: 7 cents per prospect for basic recursion, up to 50 cents for deep analysis
2. Data Source Integration
Building permits (Shovels)
CRM data (HubSpot MCP server just launched)
Custom databases (Snowflake, PostgreSQL, etc.)
3. Recursive Prompting Framework
Define data integrity rules (zero fabrication)
Map business activities to data queries
Build error handling and pivot strategies
Format outputs for downstream use
4. Deployment Platform
Clay for GTM workflows
Custom automation for enterprise scale
Integration with existing sales tools
What's Next: From PVPs to Persistent Intelligence
This is just the beginning. Jordan hinted at what's coming:
Temporal Intelligence: Agents that remember previous interactions and build on them over time.
Multi-Database Synthesis: Combining internal company data with external market intelligence.
Automated Sequences: 12-email campaigns where each message provides new, valuable insights.
The future isn't just about generating leads—it's about deploying persistent intelligence that makes every interaction more valuable than the last.
Your Complete Guide to MCP Server Implementation
Based on Friday's overwhelming response, here's what you need to know to get started:
Phase 1: Understanding MCP (Start Here)
Download Claude Desktop app
Understand the difference between local vs. cloud MCP servers
Test with a simple database connection (we'll show you how)
Phase 2: Shovels Database Access
Connect to the 170M+ building permits database
Learn the recursive prompting framework
Generate your first 10 PVPs manually
Phase 3: Scale with Clay
Build the 4-column automation workflow
Deploy across your entire target list
Measure engagement vs. traditional outreach
Let's dive into each phase.
The Bottom Line
We just demonstrated the future of B2B sales development. Recursive AI agents that turn any database into a competitive advantage, generating personalized value propositions at scale for the cost of a decent coffee.
The question isn't whether this technology will reshape GTM—it already has. The question is whether you'll deploy it before your competitors do.
As Jordan put it: "Deploy or die."
Hack the World,
Doug & Jordan
P.S. - We're building something special for paid subscribers: automated PVP generation for your business based on your company domain. Sign up with your work email, and we'll deploy recursive agents to generate personalized value propositions for your specific market. Launch coming in Mid-July Y’all.
Share this post