How to Build an Entire AI Cold Email System Live with OpenClaw & MCP

“If you’re building this today, you’re one in a million. We are very early..”

— Manthan, AI & GTM Expert

It’s 2026, and we live in the world of AI, yet most sales teams are still copy-pasting between five different tools just to send a cold email campaign. 

More tabs, not fewer. More complexity, not less. AI was meant to streamline outreach, but it’s now just another tool.

What if you could go from a raw lead list to a fully personalized, deployed cold email sequence using nothing but natural language? 

No switching between platforms, no manual CSV exports, and no copy-paste workflows that eat up your entire morning.

That’s exactly what we built in a recent live session with Manthan, using Saleshandy as the command center and OpenClaw as the AI engine. 

This blog walks you through the complete framework so you can set up the same system for your own outbound.

Let’s build it.

The Real Problem Isn’t Your Copy, It’s Your Workflow

Here’s a pattern that plays out on almost every outbound team right now. Someone discovers AI, gets excited, and starts using ChatGPT to write cold emails.

The copy gets marginally better, but pipeline doesn’t move because copy was never the actual bottleneck.

The real bottleneck is everything that surrounds the writing. 

It’s the time you spend finding the right prospects, scoring them against your ICP, personalizing beyond basic merge tags, and deploying sequences without forty-five minutes of setup every single time.

What most people call an “AI cold email system” is really just AI copy generation bolted onto the same broken workflow. You’re still bouncing between platforms and manually moving data from one tool to another.

A true AI cold email system eliminates that workflow.

You define your ICP once, set up your scoring criteria once, and build your copy framework once. Then the AI handles execution from start to finish, and Saleshandy becomes the single platform where everything runs.

Why Your Lead List Matters More Than Your Email Copy

Before we touch any AI tools, we need to address data quality because this is where most outbound campaigns quietly fail. 

The majority of teams approach lead generation with a “bigger is better” mindset, buying databases of 100,000 contacts and blasting semi-personalized templates at all of them.

This spray-and-pray approach is bad enough on its own, but AI actually makes it worse because it lets you spray faster. 

You can generate more emails in less time, which means you burn through your sender reputation quicker and hit more spam folders than inboxes.

The better approach is to start narrow and only reach out to prospects who show genuine signs of being ready to buy. 

That’s the idea behind the SPF Lead Quality Framework, a simple system for filtering your lead list down to the people who are most likely to convert.

1. S — Signal (Are They Actively Looking?)
2. P — Perfect Timing (Is the Window Open?)
3. F — Fit (Do They Match Your ICP?)

1. S — Signal (Are They Actively Looking?)

A signal is concrete evidence that a prospect needs what you’re selling right now. 

This goes beyond demographic data like job title or company size because those are static attributes that don’t tell you anything about current intent.

The signals you want to look for are actions that indicate buying behavior. 

A company hiring an SDR might be the perfect prospect for your outbound agency because they clearly need sales capacity. 

Changes in tech stack, spikes in web traffic, and new product launches are all strong signals that a company is actively investing in growth.

2. P — Perfect Timing (Is the Window Open?)

Even the perfect prospect becomes a dead lead at the wrong time.

Timing signals help you identify when someone is most likely to be open to hearing from you and considering new solutions.

Leadership changes are one of the strongest timing indicators because a new VP of Sales will evaluate every vendor within their first 90 days. 

Recent funding rounds create budget pressure that makes decision-makers more responsive to outreach. 

Geographic expansion is another powerful trigger because companies entering new markets need local infrastructure and partnerships fast.

3. F — Fit (Do They Match Your ICP?)

Your ideal customer profile shouldn’t be built on assumptions or guesswork. 

The most reliable way to define it is through pattern recognition: look at the clients who are already paying you and write down what they have in common.

You’ll almost always find shared traits in employee count, industry vertical, business model, and buying behaviour. 

Once you have that pattern, use Saleshandy’s Lead Finder to find lookalike companies that match those same characteristics.

How SPF Works in Practice

In the live session, this framework turned a massive database into a focused hit list. 

Starting with Saleshandy’s database of 800M+ contacts, the first step was entering a lookalike company URL to find similar businesses. That immediately narrowed the pool to about 72,000 companies.

After filtering for B2B companies only, the list dropped to 43,000. Adding employee count and funding round filters brought it down to a few thousand highly qualified prospects. 

The final step was using Lead Finder’s AI Search beta, where you can simply type something like “CEOs in California at Series A AI SaaS companies with 10-50 employees” and the system applies all the filters automatically.

That’s the difference between spray-and-pray and precision outbound. You’re not emailing more people, you’re emailing the right people.

How to Set Up OpenClaw for Autonomous Outreach

OpenClaw is the AI engine that powers this entire system. 

Think of it as a personal AI assistant that runs 24/7 through your phone on Telegram, WhatsApp, Discord, or whatever messaging app you prefer.

It connects to your tools through APIs, follows the instructions you’ve defined in your folder structure, and executes multi-step workflows using nothing but natural language commands.

The setup process is straightforward. 

You need a clean device (a spare laptop or Mac Mini works perfectly), a terminal to run a single install command, and a Claude API key to power the AI behind the scenes. 

The entire installation takes about ten minutes, and OpenClaw provides a step-by-step guide to walk you through it.

Once installed, you connect OpenClaw to a gateway application like Telegram. This gateway is what allows you to interact with your AI agent directly from your phone, send it tasks at any time, and receive results without needing to sit at a computer.

One important note on alternatives: everything in this blog works equally well with Claude Code if you prefer a project-based approach over an always-on assistant. 

The main difference is that OpenClaw runs continuously and maintains long-term memory across conversations, while Claude Code is session-based and better suited for focused, one-off projects. 

Choose based on whether you need an always-on assistant or a deep-focus specialist.

How to Connect Saleshandy MCP to Your AI Agent

With OpenClaw (or Claude Code) set up, the next step is connecting it to Saleshandy through MCP. MCP stands for Model Context Protocol.

It’s the bridge that lets your AI agent interact directly with your Saleshandy account, creating sequences, importing prospects, pulling analytics, and managing campaigns.

The connection takes under two minutes to set up:

1. Open your AI tool (Claude, Cursor, or similar) and go to Settings → Connectors → Custom
2. Add a new custom connector and name it “Saleshandy”
3. Paste the Saleshandy remote MCP URL (you’ll find this in the resource pack linked at the end of this blog)
4. Replace the placeholder API key with your own, which you can grab from Settings → API Key in your Saleshandy dashboard
5. Click Add Connector and refresh the page

Once connected, Saleshandy exposes 42 different tools to your AI agent.

These cover everything from creating and managing sequences to importing prospects, verifying emails, running A/B tests, and pulling campaign analytics.

To streamline multi-step workflows, go into your connector settings and select “Allow all options” so the AI can chain multiple API calls together without asking for permission at each step.

This becomes essential when you want to score leads, write emails, and deploy a sequence in a single conversation thread.

The real power shows up when you start combining these capabilities. 

You can ask your AI agent to pull last week’s campaign analytics, identify which subject lines had the highest open rates, generate new variants based on those insights, and create a fresh sequence all without leaving the chat window.

No CSV exports, no dashboard switching, and no manual data entry.

The Folder Structure That Makes AI Outreach Actually Work

This section is what separates a functional AI outreach system from one that produces generic, forgettable emails.

Without a proper folder structure, your AI agent has no context about your business, no framework for evaluating prospects, and no reference point for what good outreach actually looks like in your market.

The principle is simple: the better the context you give your AI, the better the output you get back.

Here’s the exact folder structure that powers this system, with each file explained.

1. brain.md: Your Business Context File
2. scoring-criteria.md: Your Lead Tiering System
3. copy-framework.md: Your Personalization Blueprint

1. brain.md: Your Business Context File

This is the single most important file in your entire setup. 

It contains everything your AI agent needs to understand about your business, your product, your ICP, the decision-makers you target, your value propositions, and your competitive positioning.

You don’t need to write this from scratch.

Just tell your AI tool to create a brain.md file and feed it your website URL, your pitch deck, or a verbal description of your business. 

The AI will structure everything into a clean, referenceable document that it can pull from whenever it needs business context.

There’s one non-negotiable element you must include: explicit safety rules. 

Your brain.md should clearly state that the AI must never launch a Saleshandy sequence without your direct permission. 

This single rule prevents the agent from auto-sending emails you haven’t reviewed, which is the biggest risk in any autonomous outreach system.

2. scoring-criteria.md: Your Lead Tiering System

This file defines exactly how your AI agent should evaluate and rank every prospect it encounters. A three-tier system works well for most teams.

• Tier 1 prospects are the ones that perfectly match your ICP, show active buying signals, and should be contacted immediately. 
• Tier 2 prospects match most of your criteria but might lack a strong timing signal, making them a solid secondary pipeline. 
• Tier 3 prospects don’t fit well enough to justify outreach and should be eliminated or held as backup.

When your AI agent processes a lead CSV, it scores every single company against these definitions and returns a ranked list with reasoning. 

In the live session, 25 companies went in, and only 9 came back as tier one that kind of focus is what drives reply rates 3-5x higher than untargeted outreach.

3. copy-framework.md: Your Personalization Blueprint

This file teaches your AI agent how to approach email copy for your specific market and audience.

Rather than relying on generic cold email best practices, you’re giving it a concrete reference for tone, structure, and personalization depth.

The most powerful approach is to start with one genuinely personalized email that you write manually for your single best prospect.

Research their LinkedIn profile, understand their company’s challenges, and craft an email that speaks directly to their situation.

Then feed that email into copy-framework.md as the gold standard for your AI to replicate at scale.

“Pick one prospect from your CSV. Go to their LinkedIn. Personalize everything about them. Create a copy for that one person only. Then give it to AI to reverse-engineer into a template and scale it.”

This reverse-engineering approach produces dramatically better results than starting with a template and trying to make it feel personal through merge tags.

Supporting Files

• saleshandy-api.md contains pre-scraped Saleshandy API documentation, so your agent can make API calls without burning tokens on web research every time. This speeds up execution and reduces errors significantly.
• api-keys.md stores all your API keys in one isolated file from Saleshandy, Claude, enrichment tools, and any others you use. Keeping keys separate from your business context files prevents accidental exposure.
• templates/ is a folder containing examples of your past campaigns that actually booked meetings. These give the AI concrete evidence of what “good” looks like for your specific audience, rather than having it rely on generic email templates it learned during training.

Download the full folder structure from GitHub →

From CSV to Deployed Campaign: The Step-by-Step Walkthrough

With the foundation set up, here’s how the entire system comes together in practice.

This walkthrough covers the exact steps from the live session, framed as a repeatable process that works regardless of your industry or target market.

• Step 1: Score Your Leads Automatically
• Step 2: Generate Personalized Emails at Scale
• Step 3: Deploy the Sequence in Saleshandy Automatically

Step 1: Score Your Leads Automatically

Start by exporting your focused lead list from Saleshandy Lead Finder as a CSV file. Then send it to your AI agent with a straightforward instruction: “Score these prospects using our scoring criteria and break them down by tier.”

Your agent reads the scoring-criteria.md file, evaluates every company against your tier definitions, and returns a ranked list with scores and reasoning for each entry.

In the live session, this process turned 25 raw companies into a focused hit list of 9 tier-one prospects, each with a clear explanation of why it qualified.

Here’s a cost advantage worth noting: running this same scoring and enrichment process inside Clay would cost you per API call for every single prospect.

With a Claude subscription powering your AI agent, the scoring is effectively included in your base subscription.

Step 2: Generate Personalized Emails at Scale

With your scored list ready, ask your agent to write personalized emails for each tier-one prospect. 

Because the agent has access to your brain.md for business context, your copy-framework.md for personalization approach, and your templates folder for reference campaigns, the output is genuinely tailored to each recipient.

The emails that come back aren’t template swaps with a few variable tags replaced. 

They’re structurally different for each prospect because the copy framework.md taught the agent to think like a human prospector, research the company first, identify specific pain points second, and write a unique email third.

Step 3: Deploy the Sequence in Saleshandy Automatically

This is where the MCP connection turns everything into a single seamless workflow. 

One natural language instruction: “Create a sequence in Saleshandy with these emails and import all the prospects” triggers the agent to handle everything through the API.

It creates the full sequence with multi-day cadencing across day one, day three, and day seven. 

It imports all prospects with their personalized email copies attached. And it maps every merge tag correctly without any manual intervention.

When you refresh your Saleshandy dashboard, everything is there: the sequence name, the prospect count, and all email steps with personalized copy ready to review and launch. 

No copying, no pasting, and no manual data entry anywhere in the process.

One natural language command produces one complete, ready-to-launch cold email campaign.

Running Your AI Agent 24/7 on Autopilot

Building one campaign through this system is impressive, but the real value comes when it runs continuously without your involvement. 

The goal is to have an AI agent that prospects, scores, personalizes, and deploys campaigns around the clock while you focus your time on the conversations and deals that actually need a human.

The limitation of running OpenClaw locally is simple: close your laptop, and the agent stops working.

To create a truly autonomous system, you need to deploy it to a cloud hosting platform like Railway or Hostinger so it runs continuously regardless of whether your machine is on.

Once deployed, you can set up cron jobs automated tasks that execute on a repeating schedule. 

For Example, you could have a cron job that scrapes TechCrunch every evening to find newly funded startups that match your ICP. If a startup qualifies, the agent automatically pulls their data, scores them, writes personalized emails, and creates a sequence in Saleshandy. You wake up to an email summary of everything it processed overnight.

You can also trigger outreach from real-time events.

When you find an engaging LinkedIn post in your niche, you can copy the URL, send it to your AI agent, and it will scrape all the commenters. 

It then cross-references those profiles against Saleshandy Lead Finder for verified email addresses.

It creates a targeted sequence for people who are already engaging with content relevant to your offer.

A Note on Security

Always run your AI agent on a dedicated clean device that’s separate from your personal machine.

Your brain.md file should include explicit rules: read-only access to email accounts, draft-only mode for replies (never auto-send), and a hard requirement for human approval before any Saleshandy sequence goes live.

These guardrails ensure that the agent stays productive without ever doing something you haven’t authorized.

Your Five-Day Action Plan

You don’t need to build the entire system in one sitting. Here’s a practical rollout that gets you from zero to your first AI-powered campaign within a working week.

Day 1: Lay the foundation

Create your Saleshandy account if you don’t have one yet, connect the MCP connector to Claude, and write your brain.md file.

Start by simply describing your business, your ICP, and what a dream prospect looks like.

Day 2: Build Your First Focused Lead List.

Apply the SPF framework to define your signals, timing criteria, and ICP fit. Then pull a list of 25-50 prospects from Saleshandy Lead Finder.

Resist the temptation to go broad because quality over quantity is the entire point of this system.

Day 3: Create Your Scoring Criteria and Copy Framework

Write your tier definitions in scoring-criteria.md.

Then manually personalize one email to your absolute best prospect, and save it in your copy-framework.md as the reference template for scaling.

Day 4: Run Your First AI-Powered Campaign

Feed your lead list to your AI agent and tell it to score the prospects, write personalized emails, and create a sequence in Saleshandy.

Review the output, make any adjustments that need a human eye, and activate the campaign.

Day 5: Iterate Using Real Performance Data

Pull your campaign analytics through the MCP connector and review what’s working.

Which subject lines drove opens? Which emails got replies? Feed that data back to your AI agent and have it refine the copy for your second round of outreach.

The infrastructure for this system exists today, and it’s accessible to anyone willing to spend a few hours setting it up.

The teams that build this into their GTM engine now will have a compounding advantage over the ones who wait another twelve months.

Resources and Next Steps

Everything from this session is available for you to download, fork, and customize:

• AI Cold Email Campaign Kit — GitHub Repo — Contains brain.md, scoring-criteria.md, copy-framework.md, API docs, and templates ready to adapt to your business.
• Session Slides — The complete presentation deck from the live session.
• Saleshandy MCP Setup Guide — Step-by-step documentation for connecting Saleshandy to Claude, Cursor, and other MCP-compatible tools.
• Saleshandy Lead Finder — 800M+ verified contacts with 75+ filters and AI-powered natural language search.

Frequently Asked Questions

1. Do I need coding skills to set up this system?

Not at all. Everything in this guide runs on natural language commands. The MCP connector setup is point-and-click inside Claude’s settings, and OpenClaw installs with a single terminal command.

If you can type a message in Telegram, you have the technical skills to run this system.

2. How much does this cost to run?

The Claude Max subscription costs $200/month for API access, and Saleshandy’s plans start at $25/month with unlimited email accounts.

Compared to running similar enrichment and scoring workflows in Clay, where every GPT API call adds up separately, this approach is significantly more cost-effective as you scale.

3. Can I use Claude Code instead of OpenClaw?

Absolutely. Everything in this blog works with Claude Code as well.

The main difference is that OpenClaw runs 24/7 through messaging apps and maintains long-term memory, while Claude Code is session-based and better for focused project work.

For ongoing outbound operations, OpenClaw is more practical. For one-off campaign builds, Claude Code produces excellent results.

4. What if the AI writes inaccurate emails?

It will sometimes, and that’s expected and accounted for in this workflow. The system is designed with a human review step built in before anything goes live.

AI handles the 80% of work that’s repetitive research, scoring, personalization, and formatting, while you handle the 20% that requires human judgment around tone, accuracy, and strategic decisions.

5. Is it safe to give AI access to my sending tools?

Safety depends entirely on how you configure the system. Your brain.md file should explicitly prevent the AI from launching sequences without your direct permission.

Use read-only access for email accounts and draft-only mode for replies. Running your AI agent on a dedicated device that’s separate from your personal accounts adds another important layer of protection.