🚀 AI-Powered Sales Automation: Configuring Claude Code & Instantly MCP for Instant Lead Gen

📋 Overview

• Type: Technical Tutorial / Strategic Workshop
• Main Topic: Leveraging an IDE (Cursor) and Claude Code to autonomously ingest leads, segment audiences, write cold email copy, and launch campaigns in Instantly.ai via MCP.
• Speakers: Sales Development / Agency Expert (from "Top of Funnel").

🎯 Core Purpose & Context

The goal of this session is to move beyond manual Sales Development Representative (SDR) work. The speaker demonstrates a high-leverage workflow where an AI Agent (Claude Code) acts as the operator. Instead of clicking buttons in a SaaS dashboard, the user sets up a local environment where the AI analyzes data, makes strategic decisions (segmentation), writes copy, and executes campaign creation directly via API/MCP.

---

🧠 Key Concepts & Technical Stack

Core Concepts

1. IDE (Integrated Development Environment): The workspace used. The speaker prefers Cursor over VS Code for its ability to visualize code, CSVs, and browsers in a single window.
2. R.T.O. Framework: The prompting strategy used to program the AI:
   • Role: Who is the AI? (e.g., "Mastermind Sales Developer").
   • Task: What specifically needs to be done? (e.g., "Analyze CSV and create campaigns").
   • Output: What is the deliverable? (e.g., "Active campaigns in Instantly").
3. MCP (Model Context Protocol): The technical standard that allows the AI (Claude) to "talk" to external tools (Instantly.ai) safely and structurally via API.

Figure 1: The full automation stack — Claude Code acts as the operator, using MCP as the bridge between local lead data and Instantly.ai campaign execution.

The "Rules of Engagement" for Cold Email (Defined in Prompt)

• Length: 80–100 words maximum.
• Tone: Zero small talk ("I hope this finds you well" is banned).
• Structure: High value, soft Call to Action (CTA).
• Goal: Move from cold to warm lead status; start a conversation, don't just sell.

---

⚡ Step-by-Step Implementation Guide

Phase 1: Environment Setup

1. Select IDE: Open Cursor (recommended for speed/shipping).
2. Directory Setup: Create a new empty folder on your local machine (e.g., "Claude Code Demo") and open it in the IDE.
3. Data Import: Drag and drop your .csv lead list into this folder.
4. Launch Terminal: Open the terminal within Cursor (using the VS Code plugin for Claude Code).

Phase 2: Role "Injection" (The Matrix Moment)

You must "initialize" the project by setting the context.

• Action: Use the RTO framework to tell Claude it is a "Sales Mastermind."
• Specific Instruction: "You are a professional marketer. Cut the small talk. Focus on high-value propositions."

Phase 3: Connecting the MCP (Instantly.ai)

1. Get API Key: Go to Instantly.ai -> Settings -> Integrations -> API Keys -> Create New (Give full scopes used for demo).
2. Install MCP: Run the command in the terminal: claude mcp add [transport streamable http] [URL provided by Instantly]
3. Scope: Set to "Project Level" (not global) so this API key is only accessible within this specific folder/client project.
4. Verify: Run /mcp or ask Claude "Do you have access to Instantly MCP?" to confirm tool availability.

Figure 4: The Mega-Prompt execution loop — Claude autonomously moves from raw CSV data to live segmented campaigns, pausing only for human copy approval before API deployment.

Phase 4: Execution (The Loop)

1. Prompt: asking Claude to read the CSV.
2. Analyze: Claude analyzes the leads and decides on segmentation (e.g., "Decision Makers" vs. "Influencers").
3. Create: Claude generates the emails in a Markdown file (.md) for review.
4. Deploy: Claude uses the MCP to actually create the campaigns inside the Instantly account and upload the leads to the specific campaigns.

---

🧭 Strategic Analysis & "Game Changers"

🧐 Hidden Connections

• The Death of the Drag-and-Drop Interface: The speaker implies a future where we interact less with UI (User Interfaces) and more with CLI (Command Line Interfaces) driven by natural language. We are bypassing the "Instantly" dashboard entirely until the verification stage.
• Local Privacy / Cloud Execution: By keeping the environment local (on the hard drive) but using MCP to touch the SaaS, the user maintains better control over data files (CSVs) while leveraging cloud power.

💡 The "So What?"

This workflow changes the unit of labor. An SDR typically spends hours cleaning lists, segmenting rows in Excel, writing copy, and uploading to a tool. This workflow condenses 2-3 hours of manual work into 5 minutes of prompt engineering. It allows a single person to function as an entire Lead Gen agency.

Figure 3: MCP transforms Claude from a text generator into a software operator — collapsing hours of manual SDR work into a single prompt execution.

🚀 The Game Changer: Model Context Protocol (MCP)

The true revolution here isn't the AI writing the email; it's the MCP connection.

• Before MCP: You ask ChatGPT to write an email -> Copy -> Paste into Instantly -> Upload CSV -> Map fields manually.
• With MCP: You ask Claude to "do the campaign." Claude is the software operator. It executes actions, not just text generation.

---

📊 Detailed Breakdown

[00:00:00] Introduction & Tool Selection

• Context: Code capabilities for operational tasks (not just building apps).
• Tool Choice: Cursor is selected as the IDE because it handles visualization of code and CSVs faster than VS Code or Windsurf using "Yell Mode" (term for high-velocity coding).
• The Goal: Launch a lead gen campaign using Claude Code.

[00:01:03] Project Initialization

• Setup: Create an empty folder -> Open in Cursor.
• Data Entry: A CSV of leads is present in the folder.
• Why Cursor?: It integrates the browser and terminal in one window, allowing for better "Context" (the AI sees the files and the folder structure).

Figure 2: The RTO Framework — structuring your AI prompt with a defined Role, Task, and Output is the foundation of high-conversion cold outreach automation.

[00:02:40] "RTO" - Defining the Persona

• The Prompt: The user speaks to the AI to define the Role, Task, and Output.
• The Persona: "You are a mastermind when it comes to sales development... you are a professional marketer."
• Strict Rules:
  • No "I hope this finds you well."
  • No small talk.
  • Emails must be 80-100 words.
  • Focus on "Cold to Warm" conversion.
• The Handshake: The user asks Claude to confirm understanding before executing creates a feedback loop to prevent errors.

[00:06:17] Creating the Logic (claud.md)

• Action: Claude creates a claud.md file. This acts as the "Brain" or "Constitution" for the project.
• Tech Tip: Use Shift + Command + V in Cursor to preview Markdown files beautifully.
• Strategy: "Injecting skills almost like Matrix to Neo." You are saving your strategic brain into a file that the AI references for every future task in this folder.

[00:08:00] Installing the Instantly MCP

• Process: The user grabs the MCP URL (https://mcp.instantly.ai...) and generates an API key from Instantly.
• Security Note: The speaker discusses "Project Level" vs. "User Level" installation.
  • Strategic Insight: Always use Project Level for agencies. You don't want Client A's API key active when working on Client B's folder.
• Verification: The user runs a check to ensure Instantly MCP is listed in the available tools.

[00:12:51] The Execution Prompt (The "Mega-Prompt")

• Input: The user drags sample_leads.csv into the context.
• Complex Instruction:
  1. Analyze: Look at the CSV.
  2. Score: Rate leads (Decision Maker vs. Influencer).
  3. Segment: Split them into valid groups.
  4. Copywrite: Write specific emails for those segments (referencing the Air Force case study).
  5. Execute: Create the campaigns in Instantly and push the leads there.
• Constraint: "Stop for verification." The user wants to approve the copy before the API sends it to the tool.

[00:13:37] Real-Time Automation

• Observation: Claude reads the CSV, identifies 20 leads, and splits them into "Executive Tier 1" and "Influencer."
• Action: Claude creates two separate campaigns in Instantly via the API.
• Resource Allocation: Claude asks to assign sender accounts (Email inboxes). The user specifies adding 10 accounts to the campaign.

[00:14:15] Quality Assurance

• Copy Review: The user reviews the generated email copy in the markdown preview.
• Tweaking: The user notices the sequence logic (Follow-up 1, Follow-up 2) needs a slight adjustment in the Subject Line ("Re:" vs standard).
• Final Result: The campaigns are live in the Instantly Dashboard with leads uploaded, schedules set, and copy written.

---

🔑 Key Takeaways

1. RTO is Non-Negotiable: Without setting a strong Role, Task, and Output (the "Constitution" file), the AI will generate generic, low-conversion outreach. You must "inject" your expertise first.
2. The Rise of MCP: The ability for an LLM to add leads and create campaigns via API (without the user touching the UI) is the future of admin automation.
3. Segmentation at Speed: The AI didn't just "upload" the list; it analyzed the job titles and split them into Tier 1 and Tier 2 campaigns automatically. This used to be a tedious manual CSV task.
4. Project-Level Security: When using MCPs, always confine API keys to the Project level to prevent cross-contamination of client data.
5. Visual Verification: Using a Markdown previewer (.md) is a superior way to audit AI-generated copy before verifying the API push.

❓ Unresolved Questions / Follow-up

• Cost Scaling: How many tokens does Claude Code consume when processing a CSV of 5,000+ rows? Is it cost-effective compared to traditional scripts?
• Error Handling: What happens if the CSV data is dirty (missing names, bad formatting)? Does the agent halt, or does the proper handling need to be defined in the claud.md rules?
• Sender Account Rotation: The AI added "10 accounts," but did it intelligently select which accounts based on domain health, or did it just grab the first 10?