The AI Scraping Playbook: How I Scraped 77,000 Leads in 60 Days

How I scraped 77,000 leads, booked 25 meetings, and closed 5 deals in 60 days using Claude Opus 4.7, Serper.dev, and an MCP stack most founders don’t know exists yet.

Zero Apollo. Zero Clay. Zero scraper subscriptions.

The 60-day numbers

Metric	Count
Leads scraped	77,000
Replies received	~600
Positive replies	~100
Meetings booked	25
Deals closed	5
Cash collected	$5,819
Total stack cost	~$200/month
ROI	10x+

I didn’t buy a single lead. I didn’t subscribe to Apollo. I didn’t pay Clay $500/month for credits.

I wrote a prompt. Claude Opus 4.7 wrote the scraper. Serper.dev pulled the data. Pipekit enriched it. Reachkit sent the emails.

This playbook is exactly how to build it.

Who this is for

You’re a B2B founder. SaaS, agency, consultancy. You’re either:

• Paying $500+/month for enrichment data that’s 3 months old
• Paying an agency $4,500/month to run outbound on top of your domain
• Not doing outbound at all because the stack looks impossible

This playbook tears down the middleman. You’ll learn how to scrape the exact same data Apollo sells back to you. Fresh. Live. Yours.

Fair warning: the middle of this playbook is technical. That’s the point. If you want “10 AI tools that will change your outbound,” close this tab and go to LinkedIn. If you want the actual stack that scraped 77,000 leads in 60 days, keep reading.

Why scraping broke in 2026

Three things happened in the last 18 months.

One: Apollo and Clay quietly doubled their prices. A 10,000-credit Apollo plan is now $149/month. Clay charges per enrichment on top of a seat fee that starts at $149/month per user. For a founder-led outbound setup, you’re looking at $500-$800/month just to look up leads.

Two: The data got stale. Apollo’s database refreshes quarterly. By the time you email someone, 30% of the titles are wrong, 15% of the people are at new companies, and the emails bounce like a trampoline. I once hit 12% bounce rate on an Apollo export and burned a domain.

Three: Google Maps got insanely rich. Every B2B service business now has a Google Business Profile. Website, phone, reviews, hours, category, employee count signals. All free. All live. All ignored by every enrichment provider because they can’t monetize it.

Meanwhile, Claude Opus 4.7 shipped in 2025. Claude Code lets anyone with 20 minutes of patience build a scraper that does exactly what Apollo does, at 1/50th the cost, with fresher data.

The gap between “what tools can do” and “what founders know tools can do” is the biggest alpha in outbound right now.

The stack

Four layers. Each does one job. Each hands off cleanly to the next.

Layer 1 — Research          Claude Opus 4.7 + Enrichkit MCP
Layer 2 — Scraping          Claude Code + Serper.dev
Layer 3 — Enrichment        Pipekit (free Clay alternative)
Layer 4 — Sending           Reachkit (unlimited inboxes + warmup)

Claude Opus 4.7: the brain. Does the thinking, writes the code, orchestrates the whole pipeline. Use it in two places — Claude.ai for research (with MCP connectors), and Claude Code for building the scrapers.

Enrichkit: real-time LinkedIn data. Profiles, posts, reactions, comments, company info. Connects to Claude as a custom MCP server in 2 minutes. This is the tool I built because Apollo couldn’t tell me what my prospects were actually talking about this week.

Serper.dev: Google Search and Google Maps API. $50 gets you 50,000 searches. You’ll use this to pull every B2B service business in any city or niche, with website, rating, and category attached.

Pipekit: workflow engine that chains scrapers, enrichers, and verification into a single pipeline. Free alternative to Clay. More technical, but infinitely more powerful if you can read a CSV.

Reachkit: cold email sending. Unlimited inboxes, built-in warmup, unified reply inbox. This is where your scraped list becomes booked meetings.

Total monthly cost to run this at 1,000 emails/day:

• Serper.dev: $50
• Enrichkit: $29 (Starter) or $79 (Growth)
• Reachkit: $79
• Domains + inboxes: ~$260
• Total: ~$420/month

For comparison, the same setup through an agency is $4,500+/month.

Phase 1 — ICP research (Claude + Enrichkit MCP)

Before you scrape a single lead, you need to know two things:

1. Who exactly you’re targeting (ICP)
2. What they’re complaining about this week

Most people skip this and wonder why their reply rate is 0.2%.

Why this phase is non-negotiable

A tight list with deep understanding beats a giant list with generic messaging by 10-100x. I’ve rebuilt the same campaign with the same offer, same infrastructure, only changing the list and the opening line. Reply rate went from 0.5% to 8%.

The list and the pain point you lead with are 90% of the game.

Connecting Enrichkit to Claude as an MCP server

MCP (Model Context Protocol) lets Claude use tools directly. Instead of copy-pasting LinkedIn URLs and asking Claude to “imagine what this person talks about,” you connect Enrichkit as a custom MCP server and Claude can actually look.

Claude.ai → Settings → Connectors → Add custom connector.

Enter the URL:

https://enrichkit.dev/mcp?api_key=ek_your_key

Takes 2 minutes. Claude now has access to tools like: lookup_profile, lookup_company, get_posts, get_reactions, get_comments, get_employees.

Now the fun starts.

The ICP research prompt

Open a new Claude chat. Drop this in:

You are my ICP research assistant.

I sell [your product] to [rough ICP description].

Use the Enrichkit tools to do the following:

1. Find 5 influential voices in this niche on LinkedIn.
   Start with: [3-5 names you know in the space]
2. Pull their last 10 posts each.
3. Pull the top 20 comments across all those posts.
4. Identify the top 3 pain points my ICP is actively
   complaining about this month.
5. For each pain point, quote 2-3 specific comments or
   posts that illustrate it.
6. Suggest 3 one-line opening hooks for cold emails that
   reference these pain points without sounding like a
   newsletter.

Don't summarize what I already know. Find me what's
actually on their mind right now.

What you get back is gold. Not “SaaS founders struggle with lead gen.” Actual sentences like “three agency owners in my space this week said their Clay bills went up 40% after the October pricing change, and two of them are openly looking for alternatives.”

Those are the lines you lead your emails with. That’s what “research goes behind the email, not in it” means.

What most people get wrong here

They ask Claude to “research the B2B SaaS founder ICP” without grounding it in real LinkedIn data. Claude makes up generic-sounding pain points that could apply to anyone. The emails that come out of this research are the exact emails everyone ignores.

The MCP connector is the unlock. It forces Claude to look at actual posts from actual people posting actual complaints this month.

No MCP, no research.

Phase 2 — Google Maps scraping (Claude Code + Serper.dev)

You now have a tight ICP and a handful of pain points. Time to build the list.

Most people at this stage open Apollo and filter by industry, company size, and title. That gives you a generic B2B list that everyone else is also emailing.

The alpha is Google Maps.

Why Google Maps

Every B2B service business — agencies, consultancies, home services, professional services, clinics, studios, contractors — has a Google Business Profile. It includes:

• Business name
• Website URL
• Phone number
• Address
• Category
• Review count and rating
• Years in business

This data is free, live, and structured. Apollo doesn’t sell it to you because they can’t monetize it — you can scrape it yourself with one API.

That API is Serper.dev.

Setting up Serper

Go to serper.dev → create account → grab API key. First 2,500 searches free. $50 for 50,000.

Each Google Maps search returns ~20 results. So $50 gets you roughly 1,000,000 business records. At scale you’ll pay maybe $0.0001 per lead. For context, Apollo charges you about $0.05 per record for data that’s 90 days old. You’re looking at a 500x cost reduction for fresher data.

The Claude Code workflow

Install Claude Code (Anthropic’s terminal agent):

npm install -g @anthropic-ai/claude-code

Run claude in your terminal. Drop this prompt:

I need to scrape Google Maps results via Serper.dev.

My ICP: [describe your target — e.g., "digital marketing
agencies with 10-50 employees in the United States"]

Build me a Python script that:

1. Takes a list of cities (start with top 50 US metros)
2. For each city, queries Serper.dev Maps API with:
   - query: "[niche] [city]"
   - type: "places"
   - num: 20
3. Deduplicates results by business name + address
4. Outputs to leads.csv with columns:
   business_name, website, phone, address, category,
   review_count, rating, city

Use Serper.dev API key from env var SERPER_API_KEY.
Include rate limiting (1 request per second) and a progress
bar so I can see it working.

Claude Code writes the script, saves it to disk, and offers to run it. Say yes. 50 cities × 20 results = 1,000 leads in about 3 minutes.

Want 10,000 leads? Feed it 500 cities, or run multiple niches. Want 100,000? Run overnight. The marginal cost is pennies.

Why this beats Apollo’s “small business” filter

Apollo’s small business filter is populated from LinkedIn company pages. LinkedIn company pages are filled out by maybe 40% of real businesses — the ones with a marketing person.

Google Business Profiles are filled out by 95%+ of B2B service businesses because Google forces you to claim yours to rank locally. The list you get from Google Maps is more complete, more local, and more recent than anything in Apollo’s database.

It’s also mostly founders and decision-makers, not marketing ops people.

Phase 3 — Email finding and verification (Pipekit)

You now have a list of 10,000 businesses with websites and phone numbers. What you don’t have is email addresses.

This is where Pipekit comes in.

What Pipekit does

Pipekit is a workflow engine. Think Clay, but self-hosted and essentially free. You build pipelines where each step enriches, validates, or transforms data, and the output of one step feeds the next.

For cold email list building, the pipeline looks like this:

[Import CSV of businesses]
  ↓
[Find email from Google with Serper.dev enrichment]
  ↓
[Verify email with MillionVerifier]
  ↓
[Push verified leads to Reachkit campaign]

Each step has “run conditions” that skip the lead if data is missing. So if Serper.dev can’t find an email, the lead is dropped.

Verification is non-negotiable

A 4% bounce rate burned my first domain. Under 2% bounce is the only safe zone.

Always verify before sending with MillionVerifier or NeverBounce. Costs about $0.004 per email.

If you skip this step, you will burn domains. There is no alternative.

Phase 4 — Messaging (stay human)

You have a verified list of 500-3,000 leads, tightly segmented, in your exact niche, with pain points you researched in Phase 1.

Now the simplest part. Writing the email.

The formula

Three lines. That’s the whole email.

Line 1 — About them and their pain point
Line 2 — What you do
Line 3 — Offer something small and specific

Real example:

Noticed most Austin marketing agencies are still running leads through spreadsheets and hoping referrals show up.

I set up cold email systems for agency founders so you can run outbound without hiring an SDR.

Want me to send over a 1-page cold email sample campaign that shows you how many people you can reach with this approach?

No “I hope this finds you well.” No stats. No “I noticed your industry grew 14% last quarter.” No newsletter energy.

The research from Phase 1 tells you what to put in line 1. That’s why you did it.

Why this beats AI-personalized copy in 2026

Every prospect’s inbox is full of AI emails that sound personalized but aren’t. “I saw you work at {company_name} and noticed you’re in {industry}.” This is Mad Libs, not personalization.

Your three-line email is personalized at the segment level — every marketing agency owner in Austin gets a slightly different line 1 than every accounting firm in Denver. That level of targeting requires a tight list and real research. It does not require AI rewriting every email.

AI here is used to build the list and research the pain points. Not to pretend every email was hand-written.

The math:

• 10,000 generic AI-personalized emails → 0.3% reply rate = 30 replies
• 1,000 tightly targeted, segment-written emails → 2.4% reply rate = 24 replies

Same number of replies, one-tenth the infrastructure cost, one-fifth the domain burn risk, and a list you can rerun quarterly.

Phase 5 — Sending (Reachkit)

You have the list. You have the messaging. You need to send it without burning your domains.

This is where most DIY attempts die. I watched five founders try to self-manage sending infrastructure. Four burned their domains in the first month because they skipped warmup. One burned their company’s main domain because they didn’t know not to use it.

What needs to be right

• Domains: 15-50 purpose-bought domains (never your main domain)
• Inboxes: 3 per domain maximum, 20 sends per inbox per day
• DNS: SPF, DKIM, DMARC set correctly on every domain
• Warmup: 4 weeks minimum, non-negotiable
• Rotation: sends spread across all inboxes automatically
• Reply handling: unified inbox so you can actually see responses

You can do all of this manually. It takes about 40 hours to set up and 10 hours/month to maintain. Or you can use Reachkit, which handles it end-to-end for $79/month flat regardless of inbox count.

I built Reachkit because every alternative charges per inbox, which punishes you for scaling the one variable that matters most: volume. Reachkit is unlimited inboxes on every plan.

Push your verified list into a Reachkit campaign. Set your sequence (three emails max — more is diminishing returns). Hit launch.

Reply handling

Every reply lands in the Reachkit Unibox, automatically classified — interested, meeting requested, not interested, out of office, wrong person, unsubscribe. You see what’s hot at a glance.

For the hot replies, you reply personally. That’s where the money is made. A warm reply at 9:03am handled personally at 9:04am closes at 3x the rate of the same reply answered at 2pm.

For everything else (OOO, unsubscribe, wrong person, not interested), Reachkit handles it automatically.

Could you build an AI agent that handles the hot replies too? Yes. I’ve built versions of this. For 90% of founders, the juice isn’t worth the squeeze — a warm reply answered in your own voice outperforms any agent-written reply. Use AI to build the list. Stay human on the replies.

The 4-week ramp

Don’t try to ship everything in one weekend. You’ll break something.

Week 1 — Foundation

• Sign up for Reachkit, Enrichkit, Serper.dev
• Buy 15-25 domains, set up inboxes, start warmup
• Connect Enrichkit to Claude.ai via MCP (2 minutes)
• Install Claude Code

Week 2 — Research + scraping

• Run the Phase 1 research prompt on your ICP
• Document 3 real pain points, 3 opening hooks
• Use Claude Code + Serper to scrape 5,000-10,000 leads
• Set up Pipekit pipeline for enrichment + email finding + verification

Week 3 — Launch

• Finish warmup (end of week 3)
• Write your 3-line sequence using the Phase 4 formula
• Push first 1,500 verified leads into Reachkit
• Launch at 50% volume

Week 4 — Scale

• Ramp to full volume
• Read every reply, answer hot ones personally
• Log what’s working: which pain point line gets the most replies
• Double the list, rerun the pipeline

End of week 4: system runs itself, you’re spending 20 minutes a day on replies, first meetings booked, first deals in the pipeline.

What this replaces

Put side by side, the old stack and the new stack aren’t close.

	Old stack	New stack
Lead database	Apollo ($149/mo)	Serper.dev + Google Maps ($50/mo)
Enrichment	Clay ($149/mo + credits)	Pipekit (free) + Enrichkit ($29/mo)
Research	Manual, takes days	Claude + Enrichkit MCP, 20 minutes
Sending	Instantly/Smartlead ($97+/mo)	Reachkit ($79/mo, unlimited inboxes)
Warmup	Separate tool ($50/mo)	Built into Reachkit
Reply handling	Scattered inboxes	Reachkit Unibox with classification
Agency alternative	$4,500/mo retainer	Your own stack, $200-$400/mo

The old stack treats you as a customer. The new stack treats you as a builder.

Two paths from here

Path 1 — Build it yourself

You have the playbook. Pick up Claude Code, sign up for the tools, and ship it.

• Claude Opus 4.7 — claude.ai for research, claude.com/code for building scrapers
• Enrichkit — enrichkit.dev. LinkedIn data + Claude MCP integration. From $29/month.
• Serper.dev — serper.dev. Google Search + Maps API. $50 buys you 50,000 queries.
• Pipekit — pipekit.dev. Free workflow engine for list enrichment.
• Reachkit — reachkit.ai. Unlimited inboxes, warmup, Unibox, verified sending. $79/month.

If you’re technical (or have Claude Code), you’ll have this running in a weekend.

Path 2 — Skip the build

If the middle of this playbook made your eyes glaze over, Path 2 is for you.

reachkit.ai/setup — I’ll set up the domains, inboxes, warmup, and first campaign for you. You bring the ICP, I do the rest. You own everything at the end.

The actual lesson

Every serious cold email operator in 2026 is doing some version of this.

Pulling live LinkedIn data instead of buying stale Apollo exports. Scraping Google Maps instead of paying Clay $500/month. Using Claude Opus 4.7 to do in 20 minutes what used to take a week of manual research.

The ones winning aren’t using more tools. They’re using better, cheaper, fresher tools that most founders don’t know exist yet.

The gap will close. In 12 months, every agency owner will know about Enrichkit + Serper + Claude Code. The alpha will be gone.

Right now, it’s wide open.

— Quinten

P.S. If you build this out and it works for you, email me — I read everything. If it doesn’t work and you want the shortcut, reachkit.ai/setup.