Big Idea

The argument in one line.

A complete B2B lead generation pipeline can be built in Claude Code in under 30 minutes using Firecrawl as a web-scraping MCP, replacing a workflow that previously required a paid scraper, a manual researcher, and a copywriter.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You do B2B outreach or consulting and spend hours building and researching lead lists.
You want a concrete first Claude Code project with immediate practical output.
You have a specific target audience (job title, industry, location) and want a repeatable prospecting system.
You are comfortable reviewing AI output before sending.

SKIP IF…

You need real LinkedIn data — Firecrawl cannot access LinkedIn posts or profiles due to platform restrictions.
You want a fully automated send-without-review pipeline.
You are targeting consumers rather than businesses.

TL;DR

The full version, fast.

Lead generation used to require a paid scraper, a manual researcher, and a copywriter working for days. This tutorial shows how to replace all three with three sequential Claude Code prompts and Firecrawl on the free tier. The pipeline scrapes a targeted list from business directories, enriches each contact by visiting their company website, and drafts a personalized cold email for every lead, all saved as CSV files ready to review and send. The critical discipline is starting with five leads to verify quality before scaling, reviewing every enrichment note, and saving the prompt chain as a Claude skill file so the entire workflow re-triggers with a single command next time.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:27

01 · Old lead generation workflow

Manual three-stage problem: pay for scraper, manual researcher, copywriter. A week before first send.

00:28 – 01:17

02 · System overview

Three-step system explained with on-screen labels: Find targets, Enrich each lead, Write emails.

01:18 – 02:32

03 · Tools you need

Claude Code + Firecrawl. Firecrawl homepage tour. Free tier: 1,000 credits/month, 1,000 pages.

02:33 – 03:55

04 · Set up Firecrawl connector

Add Firecrawl as custom MCP connector in Claude settings. Paste remote URL with API key, click Add.

03:56 – 05:10

05 · Test connection and generate leads

Test Firecrawl connection, then run first prompt: 20 HR directors in Austin TX. Claude scrapes in parallel.

05:11 – 06:05

06 · Review leads CSV

15 leads generated. Claude notes which had LinkedIn URLs, which did not. CSV verified in folder.

06:06 – 08:01

07 · Enrich leads data

Second prompt: visit LinkedIn or company websites; add recent_topic and enrichment_note columns. LinkedIn blocked; falls back to company sites.

08:02 – 09:47

08 · Write cold emails

Third prompt: draft personalized cold email per lead referencing enrichment note. Saved as leads_final.csv.

09:48 – 10:33

09 · Save as skill file

Ask Claude to save the full pipeline as a skill file. One-sentence trigger for all future sessions.

10:34 – 11:47

10 · Tips and compliance

Always review output. Bad personalization undermines emails. LinkedIn blocks scraping. Check ToS for any scraped site.

11:48 – 12:04

11 · Wrap up

Subscribe CTA. Weekly AI tools videos.

Atomic Insights

Lines worth screenshotting.

A week-long B2B lead gen workflow can now run in three prompts and 30 minutes of Claude Code.
LinkedIn blocks all automated scraping via Firecrawl; always build a company-website fallback into your enrichment prompt.
Bad personalization is worse than no personalization — a wrong reference destroys credibility faster than a generic email.
Start every new lead list at 5 leads, not 50 — verify output quality before burning through your scraping credits.
Claude Code shows its work in real time during multi-step scraping, so you can catch failures before they propagate to the final CSV.
A Claude skill file packages your entire prompt chain into a single-sentence trigger, turning a one-time build into a repeatable system.
Firecrawl strips messy HTML before feeding pages to Claude, dramatically cutting token cost on enrichment steps.
Match your numbers across all three prompts — if you enriched 5 leads, only 5 get personalized drafts in the email step.
Some leads will have empty enrichment data because the person does not post publicly — that is expected behavior, not a bug.
Claude flags data-quality issues mid-run (unconfirmed employers, retired leads) rather than silently filling in bad data.

Takeaway

Three prompts replace a week of manual prospecting.

WHAT TO LEARN

The same output that used to require a scraper subscription, a researcher, and a copywriter now comes from three chained CSV prompts in Claude Code.

Structure each prompt to read the previous step's CSV and write a new one — this keeps the pipeline auditable and lets you stop, verify, and restart at any stage.
LinkedIn blocks automated scraping; always include a fallback instruction in your enrichment prompt so Claude switches to company websites automatically instead of returning empty rows.
Start every new list at five leads, not fifty — catch formatting problems and bad data before burning through scraping credits.
An enrichment note that is wrong or stale is more damaging than a generic email; read every note before the email step runs.
Saving the prompt chain as a Claude skill file converts a one-time build into a single-sentence trigger for every future session, compounding the time savings across campaigns.

Glossary

Terms worth knowing.

Firecrawl: An open-source web scraping API that converts raw HTML into structured data. It connects to Claude Code as a remote MCP server and provides search, scrape, map, and crawl endpoints.
MCP connector: A Model Context Protocol integration that gives Claude Code access to external tools or APIs. Firecrawl is added as a custom MCP connector by pasting its remote server URL plus an API key.
Enrichment: The step where each raw lead gets supplementary context pulled from their LinkedIn profile or company website, used to personalize outreach emails.
Skill file: A saved prompt chain stored in Claude Code that can be triggered with a single descriptive sentence in any future session, avoiding the need to rebuild the workflow from scratch.
Credits (Firecrawl): Firecrawl usage unit. The free tier gives 1,000 credits per month, each covering one scraped page.

Resources

Things they pointed at.

01:18toolClaude Code ↗

01:30toolFirecrawl ↗

05:00linkFirecrawl MCP docs ↗

08:02linkPrompts used (Google Doc) ↗

Quotables

Lines you could clip.

10:31

“Bad personalization is gonna be worse than no personalization.”

Counterintuitive warning that reframes AI-generated emails as a liability if unchecked→ TikTok hook↗ Tweet quote

07:30

“I recommend before you go all out to start with just five and then ten.”

Practical guardrail that prevents wasted credits — specific and actionable→ newsletter pull-quote↗ Tweet quote

11:46

“That is how this can go from a one-time build to a repeatable system.”

Clean payoff line for the skill-file concept→ IG reel cold open↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Before I found Claude Code, this is what a typical lead generation setup look like for a client, especially for someone doing b to b sales or consulting outreach. You'll pay for a polo or hire someone to scrape LinkedIn manually. Then you'll export a spreadsheet and hand it over to someone to research each person manually.

Finally, you will also need a copywriter to write 50 slightly different cold emails. The whole workflow could take a week and that's before you even send a single message. Today I'm gonna show you how to build the core of that system yourself using ClaudeCode with about twenty to thirty minutes of setup.

No coding required. Here's what the system does in plain English. Step one, we tell Claude who we want to reach.

We provide the job title, industry, location, whatever that fits your target. It builds a script to pull that list from the internet or from a directory. Step two, Klot code enriches each lead.

It visits their LinkedIn or website to pull relevant contacts such as recent posts, company focus, or current roles. Step three. For each lead, it writes a personalized outreach email that references something specific about that person.

So the final output would be a spreadsheet with names, contact info, and ready to send emails. And all you have to do is review, adjust, and send it out. So that's the system.

Let's go ahead and build it. Alright. So what you'll need first is obviously clot code.

You can either run this on your computer's terminal or you can download the Clot desktop app and use it that way. The next thing that you'll need is FireCrawl. FireCrawl is what will do the actual web scraping and as you can see here, it shows you ways it can help you search, scrape, and interact with the web at scale.

It cleans up messy HTML so Clot only has to read the useful stuff. It dramatically cuts down on token cost. So for example, you could search for a website so you can see something like what they have here, top restaurants in San Francisco.

You can scrape it also so you can include a website and it will pull all the information that you need. There's also a map feature and also a crawl feature and it explains to you here at the top in the pop up that shows up what exactly it can do.

In terms of pricing, so we can see here what it is. For a free plan, you get 1,000 credits a month which is pretty nice and you can scrape 1,000 pages and then there's also a hobby plan which is $16 monthly for an annual plan, standard, and also growth.

So I'm gonna go ahead and sign up right now. Alright. So once you sign in, this is what it looks like.

It does have a lot going on but don't worry, we're only gonna use a part of this today. I'll come back to this later but for now I'm gonna go back into clot code. So I already have ClotCode connected to a folder on my computer and this is just to make sure that any files that it generates all go to the same location.

FireCrawl is not in the default list for browse connectors so you're gonna have to add it as a custom connector right here. So as you can see here, this pop up is asking for a name and a remote MCP server URL. So I'm just gonna type fire crawl and for the remote MCP server URL, you can go to fire crawl's documentation right here and I'll link this below in the description.

And here you'll see it says gives you instruction on how to run on cloud code. It provides you a remote hosted URL and it says here here's the URL that you can use.

So I'm gonna copy and paste this URL and then I'm gonna go to an empty document right here and paste it. Now you'll notice that here it's saying that you need to enter your API key. So you're gonna go back to the dashboard for FireCrawl and copy that API key and then update your link right here.

So this is what you're gonna update. And now you're gonna go ahead and copy this URL and paste it right here and press add. Alright.

So now that we did that, let's go back to ClotCode and see if this actually works. We're gonna first ask ClotCode to test its connection to FireCrawl.

Please test if you are connected to Firecrawl. Alright.

Perfect. So it's telling me that Firecrawl is connected and working. So we can go ahead and start with our first prompt.

The first prompt we're gonna use is this one right here. Use Firecrawl to find 20 HR directors at companies with 50 to 500 employees in Austin, Texas. Search LinkedIn and business directories for each lead.

Collect full name, job title, company name, company website, and LinkedIn URL if available. Save this all to a CSV file called leads dot CSV. Now obviously, is a sample prompt so update it to your liking and what you need.

So for example, if you're looking for 50 HVAC companies in, you know, Atlanta, Georgia, something like that. Go ahead and add that in.

And now we'll let Claude do its work. And I didn't mention this earlier but if this is your first time using Claude, you'll notice pop ups like this throughout your session. That's fine.

That's just Claude asking you for permission to run different task. So I usually just like always allow in this project and you just have to keep selecting that throughout your session. Alright.

So as you can see here, ClotCode is still working but I wanted to show some of the updates on what work it has been doing. Number one, says I'll run multiple searches in parallel across LinkedIn, business directories, and people search sites to gather 20 leads. Good start.

I've several leads already. Let me run more targeted searches and process the large result files in parallel. So that's one thing cool about Clot.

You can see what it's doing real time. Alright. So it was taking a while to generate 20 leads so I asked it to stop at 15 so that we can get the video moving.

As you can see here, it has generated a list of people for me that I can reach out to that fit the criteria that I asked for in my previous prompt. It also provided me some notes right here about who had LinkedIn URLs and who did not. For those that did not have LinkedIn URLs, it pointed me to a company website instead.

It also provided some specific details such as retirement on certain people. And now to double check whether the CSV file was made, so I'm gonna go ahead and go to the folder that I created and yes, I can see it right here. So that spreadsheet was created and this is nice because I can email that out to someone else to take over or just keep it on my computer.

And it was named leads dot CSV just like I requested. And when I open it up, it looks like this which is pretty nice. Alright.

So now let's go back into clot code. This next thing we're gonna do is use this prompt right here, Read leads.csv for each lead that has a LinkedIn URL, visit their profile using fire crawl and find the most recent post or activity.

Add two columns to the CSV, recent underscore topic which would be a five to 10 word summary of what they've been posting about and enrichment underscore note which is one sentence I could reference in an outreach email. Save this as leads underscore enrich dot CSV. If LinkedIn is blocked, and I meant to say blocked right here, switch to using company websites instead.

Okay. So let's go ahead and press enter. Now actually one thing I do want to do, I wanna make sure it's coming out how I would like it to be.

So I'm just gonna say do this for the first five leads. So that way it's turning out to be the results that I want and you're not simply blowing through tokens. And I recommend that for you also before you have like a 100 list of leads that you're looking for.

I recommend before you go all out to start with just five and then 10 and stuff like that. So for the sake of this demo, I'm just gonna do the five leads right here. Alright.

So it is done right now. Unfortunately, LinkedIn is fully blocked by fire crawl. So what ClotCode did was switch to company websites as planned.

And as you can see here, it provided some key findings per lead. For example, with the first lead, the company just launched an AI compliance tool. The second lead studio is celebrating fifteen plus years, recently acquired a Colombian studio and actively staffing.

But you also notice that some of them might say something like no public activity and that's normal because that person just doesn't post much probably. So in that case in your outreach email, it's gonna be more basic. And Clothcodes not gonna be 100% perfect.

Gonna make mistakes. It might not pull up all 20 leads that you asked for or maybe there might be some empty rows. So in that case, you will just follow-up with a prompt saying like, hey, row two for this lead is missing, something like that.

Please fix that. And now we're gonna use this third prompt right here to write our outreach email. Read leads underscore enrich dot c s v.

For each lead, write a short cold outreach email. Number one, open with a specific reference to their recent topic or Richmond note, if any. Number two, introduce me as and insert your name right here.

Who helps mid sized companies streamline HR operations using AI tools. Again, you're gonna update this prompt with your specific information. Number three, end with one low commitment.

Ask a fifteen minute call or a reply to learn more. Add the email as a new column called draft underscore email and save it as leads underscore final dot c s v. One thing I do wanna point out is notice how each prompt stack on top of each other and that's part of organizing your documents really well to find what you need.

So let's go ahead and press enter. Alright. So clot code is done.

It has created a leads underscore final CSV file for me. One thing I do wanna point out, remember how with the previous prompt, I only asked ClotCode to add personalized hooks for the first five? Because of that, it was only able to provide personalized information for the first five emails that it drafted in this prompt.

So just make sure your numbers are always matching with all the prompts. If you have 20 leads, make sure that you're using twenty twenty twenty for all the prompts that you have. Alright.

So let's see the documents that it created for me. As you can see here, it has a leads.csv which has all the contact information for each lead.

We have leads underscore enrich which we only did for the first five in this demo that provided specific updates on them. And finally, leads underscore final takes everything together and puts them into one spreadsheet that includes the full name, job title, company name, company website, LinkedIn URL, recent topics, enrichment note, and the draft email.

Alright. So the last thing you wanna do is create something called a skill file. This is basically just instructions that you can provide to Claude again without having to copy and paste instructions over and over.

So what I did was I just said I would like to use all of this again. Please save this as a skill file for future reference. So as you can see here, the skill file has been packaged.

It's saying that it has saved it as lead gen HR skill and it was saved into my lead generation folder on my desktop. Claude also mentioned that it was installed and active in my Claude skills library so I can invoke it immediately for any future use by describing what I want. And Claude is saying to reuse it again, just say something along the lines of find 20 HR directors at 50 to 500 employee companies in Dallas.

The skill will kick in automatically and run the full pipeline. Alright. So now for some pointers and tips that I myself learned throughout this process.

As you can see, ClotCode is good at this but definitely not 100% perfect. This is still not gonna replace you reviewing it and looking through all the leads and making sure they are your target customers. You'll also wanna make sure to check the enrichment notes and read every single email.

A bad personalization is gonna be worse than no personalization. If you don't think you're going to review well, just skip the personalization part and just ask for the cold outreach emails.

If you reference something wrong, it's gonna undermine the whole email to the lead that you want. Number two is LinkedIn scraping. As you can see, Firecrawl was not able to access LinkedIn information unfortunately.

Unfortunately, I looked into it a little bit more and it looks like LinkedIn prohibits automatic scrapping. I might be wrong though so if you know of a tool that can do that, feel free to comment below, which goes to my next point which is platform rules. Make sure to check the terms and service for any website that you're scraping.

And finally, once you've dialed in on the prompts that you like for the lead generation system that you have, make sure to save them as a clot code skill like how we did. This ensures you can trigger the whole workflow with a single command the next time you need it.

That's how this can go from a one time build to a repeatable system. And that's all for today. We were able to scrape a targeted list, enrich it with real context, draft personalized outreach emails all in Claude code.

If you found this video helpful, please subscribe to my channel. I post weekly videos on AI tools and how to use them. See you in the next one.

The Hook

The bait, then the rug-pull.

Hiring a scraper, farming out manual research, paying a copywriter to write fifty slightly-different cold emails — that was the standard B2B lead pipeline, and it burned a week before a single message went out. This breakdown walks through a three-prompt Claude Code workflow that replaces the entire stack, from targeted list to personalized draft, using only Firecrawl's free tier.

Frameworks