Section 1
What You're Actually Setting Up
ScrapeGraphAI is an open-source Python library with almost 30,000 stars on GitHub. It uses an LLM to read a webpage and pull out exactly the structured data you ask for, instead of you writing scraping code by hand. That's the tool everyone's been reposting.
The part most people miss: the same team also ships an official MCP server, which is what actually lets Claude use it directly, in plain English, with zero Python. That's the piece this guide connects. Once it's wired up, you stop opening a dashboard or writing a script. You just tell Claude what kind of leads you want, and it goes and builds the list.
What you'll have when you finish this guide: ScrapeGraphAI connected to Claude, a free API key, the exact prompt that turns any search into a clean lead list, how to scale that to hundreds of leads from one directory site, and how to have Claude run the whole thing automatically every morning.
Section 2
Get Your Free API Key (1 Minute)
ScrapeGraphAI runs on credits. Every new account gets 500 free credits, no subscription needed to start. A single lead-list search costs about 30 credits, so you get roughly 16 free runs before you'd ever need to pay a cent.
Sign up
Go to scrapegraphai.com/signup. Your 500 free credits land immediately.
Grab your API key
Head to scrapegraphai.com/dashboard and copy your key. You'll paste it in during Section 3, but keep this tab open, don't paste it into a chat with Claude directly.
Section 3
The "Install This For Me" Prompt
Open Claude Code in your terminal (or Claude Desktop with file access) and paste this. It has Claude run the official installer, ask you for your key at the right moment, and confirm the connection actually works before moving on.
Copy-paste this into Claude
Install the ScrapeGraphAI MCP server (github.com/ScrapeGraphAI/scrapegraph-mcp) so you can search the web and build me structured lead lists. 1. Run this in my terminal: npx -y @smithery/cli install @ScrapeGraphAI/scrapegraph-mcp --client claude 2. I have a free ScrapeGraphAI account with an API key from scrapegraphai.com/dashboard. When the installer asks for it, tell me exactly where to paste it, don't ask me to paste it into this chat. 3. Once it's installed, restart and confirm the connection is live by running the "sitemap" tool against https://zachdoesai.com and telling me how many pages it found. Walk me through each step in plain English as you go.
Why this works: npx -y @smithery/cli install @ScrapeGraphAI/scrapegraph-mcp --client claude is the official installer straight from ScrapeGraphAI's own MCP repo. It handles the config file for you. The only manual step left is pasting your key somewhere Claude tells you to, not into the conversation, so it never ends up logged anywhere it shouldn't.
Section 4
The Exact Prompt That Builds Your Lead List
This runs the searchscraper tool, which searches the open web and pulls structured data out of whatever it finds. Swap the bracketed business type and city for whatever you're targeting.
Copy-paste this into Claude
Using the ScrapeGraphAI MCP, run the "searchscraper" tool with num_results set to 10. user_prompt: "Find chiropractors in Austin, TX. For each one, get the business name, website, phone number, email address if listed, and physical address." Once it returns, take everything it found and: 1. Put it in a table: Business Name, Website, Phone, Email, Address 2. Mark any row missing a phone or email as "needs manual follow-up" 3. Save the table as a CSV at ~/Desktop/leads-austin-chiropractors.csv Then tell me how many usable leads (has both a phone or email) you found.
Swap the business type and city, run it again. One run against 10 results costs about 100 credits. Start with num_results: 5 the first time you try a new niche, see if the data quality is what you want, then scale up.
Section 5
Scaling to Hundreds of Leads
Section 4 is great for a targeted, one-off list. When you want hundreds of leads at once, point Claude at one directory instead of the whole web, something like a Chamber of Commerce member directory, a trade association's "find a member" page, or an industry listing site. Then have it crawl the entire thing.
Copy-paste this into Claude
Using the ScrapeGraphAI MCP, first run "smartcrawler_initiate" on [DIRECTORY URL] with extraction_mode set to "markdown", same_domain_only set to true, and max_pages set to 100. This just maps the site cheaply first. Poll "smartcrawler_fetch_results" with the returned request_id every 30 seconds until status is "completed." Look at what came back. If the listing pages clearly contain business name, contact info, and address, run "smartcrawler_initiate" again on the same URL, same settings, but extraction_mode set to "ai" this time, with this prompt: "Extract business name, website, phone number, email, and physical address from this listing page." Poll for results the same way, then combine everything into one CSV at ~/Desktop/leads-directory.csv, deduped by business name.
Why the two-pass approach: markdown mode costs 2 credits a page, AI mode costs 10. Mapping the site first for 200 credits before committing to a 1,000-credit AI pass means you never burn a big chunk of your balance on a directory that turns out to be structured in a way the crawler can't parse cleanly.
Section 6
Set It to Run Every Morning
Everything above works fine run by hand. But the real unlock is having Claude run it on a schedule, so fresh leads are sitting there before you've had coffee, with zero effort from you after setup.
Claude has a built-in scheduling feature for exactly this. If you haven't set one up before, my 3 Hidden Claude Features guide walks through the click-by-click setup in about 2 minutes. Once you're there:
Create a new scheduled routine
Paste in the exact Section 4 prompt, business type and city already filled in.
Add a dedupe instruction
Tell it to check any new results against yesterday's CSV first and only append leads that aren't already on the list. Otherwise you'll get the same 10 businesses every morning.
Set it to daily, 7am
100 credits a day is well inside a $20/mo plan if you go past the free 500, and most people never need to run it daily forever, just long enough to build out a market.
Section 7
Staying on the Free Tier
500 free credits go fast if you're not paying attention. Three rules to stretch them:
Test small before you scale
Run num_results: 5 on a new niche or city first. Confirm the data's actually good before you spend the credits on 50 results.
Map before you crawl
Always markdown mode first on a new directory (Section 5). It's 5x cheaper than AI mode and tells you whether the site is even worth the full pass.
Check your balance weekly
Open scrapegraphai.com/dashboard after your first few runs so you know exactly what each search costs before you're relying on it daily.
Section 8
Where I'd Start
Run Section 3 today, that's the whole install. Then run Section 4 once against whatever type of customer you're actually trying to reach right now, not a hypothetical example. Ten results, real business type, real city.
If the list comes back clean, that's the moment to decide: is this a one-time list, or is this a market I want fresh leads from every single day? If it's the second one, Section 6 takes five more minutes and you never have to think about it again.
That's the whole pitch. Everyone's still buying lead lists or paying by the scrape. This is the same result, free to start, and it runs itself.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.