🚀 MASTERING CLAUDE CODE & MCP: The Ultimate Guide to Automating Outbound Campaigns in Cursor

📋 Overview

• Type: Tutorial / Technical Walkthrough
• Main Topic: Utilizing Claude Code (via Cursor IDE) and the Model Context Protocol (MCP) to automate the setup, segmentation, and copywriting of cold email campaigns in Instantly.ai.
• Speakers: Technical Marketer / Sales Developer (Top of Funnel Team).

🎯 Core Purpose & Context

The video aims to bridge the gap between software development and marketing execution. The speaker demonstrates how to move beyond using AI as a text generator and instead use it as an agentic operator. By setting up a specific environment (Cursor) and connecting it to a third-party tool (Instantly) via an MCP server, the goal is to radically reduce the time spent on administrative sales tasks (list cleaning, writing, campaign setup) from hours to minutes using natural language commands.

---

🧠 Key Concepts & Technical Definitions

1. The Tech Stack

• IDE (Integrated Development Environment): The software used to write code. The speaker prefers Cursor over VS Code or Windsurf because it visualizes code and CSVs better and integrates the browser within one window.
• Claude Code / Quad: The AI agent running inside the terminal/IDE. It creates files, executes terminal commands, and reasons through tasks.
• MCP (Model Context Protocol): A standard that allows AI models to interact with external data and tools. In this video, an "Instantly MCP" allows Claude to talk directly to the Instantly.ai API to create campaigns and upload leads.
• RTO Framework: A prompting strategy used to define the AI's behavior:
  • R (Role): Who the AI is (e.g., Expert Sales Developer).
  • T (Task): What needs to be done (Initialize project, analyze leads).
  • O (Output): How the result should look (Markdown files, completed campaigns).

Figure 1: The four-component tech stack — Cursor, Claude Code, MCP, and Instantly.ai — and how data flows between them.

2. Step-by-Step Execution Guide

1. Environment Setup: Open a new empty folder in Cursor.
2. Initialize Agent: Open the terminal (Claude Code/Quad) and ingest current files.
3. Define RTO: Use a robust prompt to establish the AI's persona as a "World-class Sales Developer" (No fluff, high value).
4. Install MCP Server:
   • Get API Key from Instantly (Settings > Integrations).
   • Run command: quad mcp add using the specific transport URL (streamable http).
   • Crucial Step: Set to Project Level (not Global) to keep API keys isolated to this specific client/project.
5. Data Ingestion: Drag and drop the CSV (leads list) into the project folder.
6. Analyze & Segment: Instruct AI to read the CSV, analyze decision-maker levels (Tier 1 vs. Tier 2), and segment them.
7. Execution: Instruct AI to write copy based on the segment, create the campaigns in Instantly, and upload the leads.

Figure 4: The paradigm shift from AI-as-chatbot to AI-as-operator — humans move from 'doer' to 'verifier.'

---

🧭 Strategic Analysis & "Game Changers"

⚡ The "So What?": From Chatbot to Operator

This video illustrates the pivotal shift occurring in AI adoption. Most users interactions with AI are passive (Give me text -> Received text -> Human pastes text into tool). The Paradigm Shift: The speaker is using AI to operate software. The AI isn't just writing the email; it is logging into the CRM, setting up the campaign, configuring the delays, and uploading the contacts. This moves the human from "doer" to "verifier."

🔗 Hidden Connection: The "Vibe Coding" Revolution

The speaker mentions "vibe coding." This refers to building software or workflows without deep knowledge of syntax, relying instead on natural language and intent. Implication: Marketing teams can now build their own internal automation tools without waiting for engineering resources. The barrier to entry for complex API integrations has effectively dropped to zero using tools like Cursor + MCP.

Figure 3: Project-level MCP isolation prevents cross-client API key exposure — critical for agencies managing multiple accounts.

🏆 The Game Changer: Project-Level MCP Architecture

The speaker emphasizes installing the MCP server at the Project Level rather than the User/Global Level.

• Why this matters: For agencies managing 10+ clients, you cannot have one global AI agent with access to every client's API keys. By isolating MCPs to specific folders (projects), you create secure, containerized AI environments. This enables scalable agency operations where each client folder has its own "Brain" and "Tools."

---

📊 Detailed Breakdown

[00:00:00] Introduction & Tool Selection

• Context: Cloud/Claude code transforms how we build apps and manage workflows.
• Tool Choice: Recommendation to use Cursor (IDE).
• Why Cursor? It excels at visualization (ships faster), allows viewing CSVs and the code in one window, and keeps the browser integrated.
• Alternative Agents: Mentions Windsurf, Zed, and standard VS Code, but Cursor is the current "king" for this workflow.

[00:01:06] Project Initialization

• Measurement: Starts with an empty directory.
• Action: Create a new folder (e.g., "Cloud Code Demo").
• Data Import: Dragging a CSV file of leads into the folder.
• Interface: Using the VS Code plugin for Claude code within Cursor for a cleaner UI ("Yellow Mode") vs. raw terminal.

Figure 2: The RTO Framework — defining Role, Task, and Output is the difference between generic AI output and precision execution.

[00:02:44] The RTO Prompting Strategy (Crucial Segment)

• RTO Definition: Role, Task, Output.
• Role Prompt (Verbatim Nuances):
  • "You are a mastermind when it comes to sales development."
  • "Cut the small talk. No 'Hope this finds you well'."
  • "Nobody cares how shiny your LinkedIn profile is."
  • Length Constraint: 80-100 words max.
  • Tone: High value, soft call to action (CTA).
  • Goal: Move from "Cold Lead" to "Warm Lead" (not straight to close).
• Verification: The user asks Claude to confirm understanding and propose steps before executing. This is a critical safety mechanism in AI prompting.

[00:05:54] Documentation & Setup

• Outcome: Claude creates a quad.md (Markdown) file to document its plan.
• Git Init: Claude automatically attempts to initialize a Git repository. The speaker accepts this ("Yes to all") to keep version control of the campaign logic.
• The "Skill Injection": The speaker compares the RTO prompt to "The Matrix"—injecting Kung Fu skills into Neo. You are effectively "cloning" your best strategist into the specific project folder.

[00:07:00] Connecting the Instantly MCP

• Security: Speaker navigates to https://mcp.instantly.ai.
• API Key Creation: Goes to Instantly > Settings > Integrations > API Keys.
• Command Execution: quad mcp add + Transport Streamable HTTP + API Key URL.
• Strategic Detail: The speaker explicitly instructs Claude to add this at the Project Level.
  • Quote: "I don't want to install this MCP server with that API key... for all of my projects."

[00:10:00] Data Analysis & Segmentation

• Input: The CSV (sample leads.csv) contains raw data.
• Process: AI is instructed to:
  1. Analyze the list.
  2. Score leads based on "Decision Maker Level."
  3. Segment them into appropriate tiers (Executive vs. Influencer).
• Speed: The AI rips through the analysis in seconds, identifying Tier 1, Tier 2, and Tier 3 leads automatically.

[00:12:00] Copywriting & Context Injection

• Context Given:
  • "We help teams save time, money, and raise top line."
  • "Case studies: US Air Force, Mid-market clients."
  • CTA: "Video on how this works" or "Personalized proposal."
• Execution: Claude writes the email copy and saves it to a markdown file for review (Shift + Command + V to preview in Cursor).

[00:13:00] Campaign Deployment (The "Magic" Moment)

• Action: Claude is instructed to:
  1. Create two separate campaigns in Instantly (based on the previous segmentation).
  2. Assign sender accounts (Assigns 10 accounts from a pool of 67 eligible).
  3. Upload the specific leads to their respective campaigns.
• Result: "Both campaigns have been created successfully."

[00:14:00] Verification & Polish

• Human-in-the-Loop: The speaker manually logs into Instantly to verify.
  • Status: Campaigns created.
  • Sequences: Emails are populated.
  • Leads: Contacts are uploaded.
• Cleanup: The speaker notes a small formatting error in the signature/spacing and manually tweaks it.
  • Observation: AI is 95% there, but the final 5% formatting check is essential for professional outreach.

[00:16:11] Conclusion

• Summary: In under a few minutes, the user went from a CSV file to two fully segmented, written, and configured campaigns without manually clicking through the Instantly UI.

---

🔑 Key Takeaways

1. AI as Middleware: The most powerful use of LLMs today is not content generation, but acting as the middleware between data (CSV) and platforms (Instantly) via MCP.
2. The "RTO" Framework is Mandatory: Without defining Role, Task, and Output, the AI will default to generic, low-quality "sales" language. You must program the constraints (e.g., "No small talk").
3. Cursor is the Marketer's New Dashboard: For technical marketers, Cursor is replacing the browser as the primary workspace because it unifies file management, code execution, and AI reasoning.
4. Segmentation at Scale: You can use LLMs to "score" and "segment" leads based on job titles inside a CSV before importing them—something that usually requires complex spreadsheet formulas or manual tagging.
5. Project-Level Isolation: For security and organization, always install MCP tools at the project level, especially when dealing with client API keys.

❓ Unresolved Questions / Follow-up

• Cost of Token Usage: Deep analysis of heavy CSV files using Opus 4.5 (which the speaker mentions) can be expensive. What is the cost-benefit ratio for larger lists (e.g., 10,000 leads)?
• Error Handling: What happens if the CSV data is messy? Does the MCP connection fail gracefully or upload bad data?
• MCP Source: The speaker mentions "I will provide [the MCP] to you." It is unclear if this is a public Instantly MCP or a proprietary one built by his team/Instantly. (Note: He types mcp.instantly.ai, implying a public endpoint).