Blog · AI use cases

How to Build a LinkedIn Job Scraper That Finds Hot Leads for Your Agency (with Claude Code)

A step-by-step build of an internal tool that finds companies hiring for SEO on LinkedIn, then unlocks the owner's name and email with Apollo. Built with Claude Code, no coding.

By Vai S. · 26 Oct 2026

This is what we built. It finds you hot leads.

Let’s say you’re an SEO agency. You can look for companies that are hiring for an SEO specialist in the US and posted that job in the last one week. If a company is hiring for SEO, it already has a budget for SEO. That’s your signal.

SEO Job Prospector internal tool showing LinkedIn SEO job posts with an Unlock button for each company owner

Click on the job title and you can see the job description. If you like the company and want to approach its owner, click on Unlock. It unlocks the owner’s name and email ID, and now your team can reach out.

In this post, I’ll show you how I vibe coded this tool with Claude Code, step by step. You’ll see me building it, watching it build, running it, finding bugs and adding a new feature.

Prefer to read? Here’s every step with a short clip.

What we are building

We are going to turn a signal into a lead.

Slide: the signal is a company posting an SEO job on LinkedIn, the lead is the owner of that company with contact details

The internal app fetches LinkedIn job posts that mention SEO and are located in a specific country. I’m using the US for this example. The output is the company owner’s name and email ID.

This is the process:

  1. Your team enters the keyword and the number of job posts to fetch.
  2. Apify fetches the matching job posts.
  3. Apollo finds the company’s owner and their email ID.
  4. Your team reaches out to the owner.
Four-step flow: your team enters keywords, Apify fetches matching job posts, Apollo finds the company owner, your team reaches out

Keep the cost low

Apify gives you only $5 a month if you don’t want to go into the paid plan. So we keep the cost low with two controls:

  • Keyword entry, so we only pull the jobs we care about.
  • Fetch control, so we decide how many job posts to pull. Of course, we don’t want to fetch 1,000 job posts. Let’s say 25 or 50 to begin with.
Slide: keyword entry and fetch control keep results relevant and the Apify bill low

Apollo is the same. It only unlocks an owner when you click the button, because I don’t want to burn my Apollo credits on companies I’ll never contact.

Step 1: Install VS Code and Claude Code

The tool we are going to use is VS Code. It’s free. Just Google “download VS Code”, click on the first link and download it for your PC or Mac.

Download Visual Studio Code page with Windows, macOS and Linux options

Once it’s installed, click on the Extensions button and search for “Claude”. Install Claude Code for VS Code. For me it’s already installed, that’s why it shows Uninstall. For you it will show Install.

Claude Code for VS Code extension page in the VS Code marketplace

After installing, you’ll see a Claude button at the top. Click on it and Claude opens right there inside VS Code.

Step 2: Copy the prompt

Now we are going to use this prompt. I’ve written a very detailed one: the system requirements, the reasoning or logic flow, the process and the output I want.

The LinkedIn job scraper prompt page on efficialabs.com

I want the output in a table like this: date posted, job post title, company, owner name and owner email ID. And it only unlocks the owner when I click the Apollo button.

At the end, there are some important considerations and reminders. Look at the last line. I added an instruction that says: before you start, ask me relevant questions until you understand what I want, then summarize what you understood.

Selecting the full prompt, including the important considerations and the ask-me-questions reminder

This one line makes a big difference. Claude doesn’t guess. It asks.

Step 3: Paste it into Claude Code in plan mode

Copy the prompt and paste it into Claude in VS Code. Before sending it, change three things:

  • Model: Opus 5.5
  • Reasoning: Medium
  • Plan mode: on
The prompt pasted into Claude Code inside VS Code

In plan mode, Claude will just plan the internal app for us. It won’t build the app right away.

Step 4: Answer Claude’s questions

Now Claude asks me some relevant questions to make sure it understands what I want.

Who is the “owner”? The LinkedIn job poster, or the company decision-maker? Company decision-maker only, yes. It also set the title priority for Apollo: Founder and CEO first, then Owner, CMO and Head of Marketing. One best match per company.

Hosting? I want it local only. I don’t want to deploy it to any server, so it runs only on my local machine.

Storage? It offered Google Sheets, but that makes this project a bit more complex because I’d need the Google Cloud API too. Let’s keep it session only, because I’m making this tool only for this video.

Claude asks whether fetched jobs and unlocked contacts should be saved between sessions

Filters? Location, date posted and work type. Do I want any more? I don’t think I need more right now. If I do, I’ll just ask Claude to make the changes.

Claude asks which search filters to add besides keywords and max jobs

While asking, it also fetched my official website for the brand guidelines, because I mentioned in the prompt that the tool should look like my website.

Step 5: Review the plan

After asking me questions two times, Claude built a plan for me to review.

Claude's implementation plan for the LinkedIn Job Scraper with Apollo Unlock, including what it understood from the Q&A

It even pulled the brand tokens from my website: the deep green, the navy, the amber for the Unlock button, the Poppins and Outfit fonts. That’s why the final tool looks like EfficiaLabs.

Plan section showing the tech stack and brand tokens taken from the efficialabs.com Elementor kit

I’m not worried about the tech stack right now. First, it’s just a YouTube video, and I’ll dump this tool after posting. Second, I’m not building it for my clients. I just want it to work so I can show you the output.

I went through the step-by-step workflow. Okay. Okay. I like this plan. Click on Yes, and use auto mode.

Step-by-step workflow in the plan with the Accept this plan options

Step 6: Let Claude cook

Now Claude starts building the app.

Claude Code writing the app files, layout.tsx and page.tsx, in VS Code

It’s really interesting how good AI is at coding nowadays. If someone had told me back in 2019 that AI could code this well, I would have laughed.

It took roughly 15 minutes. When it was done, it created an .env.local file where the API keys go. Never share this file and never commit it.

The .env.local file with empty slots for the Apify API token and the Apollo API key

Claude also told me what it checked and where the build differs from the plan. One important note: the Apollo key must be a master key, otherwise the people search gets refused.

Claude's build summary: what it checked, how to start, and where the build differs from the plan

Step 7: Add your API keys

I forgot to tell you what Apify is. Apify is nothing but a scraper. It can scrape almost anything: Google Maps listings, Instagram posts of a particular profile, Instagram comments, LinkedIn comments, LinkedIn job posts.

Apify homepage, a marketplace of ready-to-run scrapers

To get your API key, click on Go to Console, then Settings, then API & Integrations. There you’ll see your API token. Copy it.

Apify console, Settings, API and Integrations page with the personal API token (blurred)

Do not share your API key with anyone. It’s a secret. That’s also why I paused the recording while pasting my keys.

Paste the Apify token and the Apollo key into .env.local. Then tell Claude: “I’ve pasted both API keys.” Let it do some checks.

Step 8: Run it

Copy the localhost link, open Chrome and paste it. Yeah, it’s working.

SEO Job Prospector running on localhost with keyword, location, date posted, work type and max jobs fields

Keyword: SEO Specialist. Country: United States. Date posted: past week. Work type: doesn’t matter. Max jobs: 25. Click Fetch jobs.

The tool fetching LinkedIn jobs through Apify, with loading rows in the table

Give it some time. It was taking more time than I anticipated, so I asked Claude. It ran a test fetch itself and showed me what came back.

Claude's test run results: jobs found, company websites and the result of each Unlock (names blurred)

I kind of like the UI, by the way. Everything is so clean, and it matches our website.

Step 9: Find your ICP and unlock the owner

It worked. Click on any job title and you can see the full job post on LinkedIn.

A LinkedIn job post for an SEO and AI Search Specialist opened from the tool

Now, don’t unlock every row. If the company falls under your ICP, then unlock it. Let me find a company that’s small enough.

Blacksmith Agency. I know this won’t fall under your ICP, but it falls under mine, because we serve advertising agencies. This is what we look for: an agency with 11 to 50 employees.

Checking Blacksmith Agency's LinkedIn page: an advertising agency with 11 to 50 employees

Click on Unlock. Decision-maker found, but no email ID for them. Interesting.

Unlock result: decision-maker found, but Apollo has no email for them

Apollo couldn’t find the decision-maker’s email for a few of these companies. But for one company it found the CEO, with his name and email ID. So this is how it works.

An unlocked row showing the company CEO's name and email (blurred)

One honest note: this is a problem with Apollo’s data, not with the tool. Some companies just don’t have an email on file. The good part is that no credit is spent when there’s no email.

Step 10: Fix the bugs

I found a few bugs:

  1. Some rows were repeated twice.
  2. The column width changed when I unlocked a decision-maker. It looked so cluttered.

I took screenshots, pasted them into Claude and asked: “I found a few bugs. Can you please fix it?”

Reporting the duplicate rows and column width bugs to Claude with screenshots

Let Claude cook for some time. Reload, fetch jobs again, unlock. See, now it’s not moving.

After the fix: fixed column widths and no duplicate rows after unlocking (name and email blurred)

Step 11: Add a new feature, daily auto-run

Remember I told you we’d be adding new features? While it was unlocking, I thought of one: make it auto-run every day at 9 AM, so I don’t have to open this internal tool and run it manually.

I told Claude: “I want to add one more feature: scheduling an auto-run at 9 AM for now, but keep the time flexible so I can change it later in the user interface.”

Asking Claude Code to add a daily auto-run at 9 AM with a flexible time

By the way, I use Wispr Flow for speech-to-text. It’s much easier and faster than typing.

Claude pushed back a bit. It said scheduling doesn’t fit because the app runs locally. Fair point. I told it: just add the feature now. It doesn’t matter if it works right now. I’ll move it to Cloudflare later.

Then it asked a few more questions. Auto-unlock? No. Keep unlocking manual. Otherwise it would burn my Apollo credits in a single day.

Claude's follow-up questions about where scheduled results go and what the 9 AM run should search for

Cool. Now we have a daily auto-run. It runs at 9 AM every day with the keyword SEO Specialist, US, past week, 25 jobs.

The Daily auto-run panel set to 09:00 IST with the saved search

One thing to know: because I’m running this on my local machine, the auto-run won’t fire when I turn off the server. It only works once it’s hosted somewhere, like Cloudflare or any server. Once it’s deployed, you never need to run it manually again. It just runs daily.

Quick recap

  1. Install VS Code and the Claude Code extension.
  2. Get the prompt, paste it in plan mode and answer Claude’s questions.
  3. Review the plan, then let Claude build it.
  4. Add your Apify and Apollo keys to .env.local.
  5. Fetch jobs, check the company against your ICP, then unlock the owner.
  6. Report bugs with screenshots and add features the same way.

Want this built for your agency?

I build agents like this with my co-founder, for agencies. This LinkedIn job scraper is one of the tools we custom-build for our clients. For example, we built an agent that cut weekly client reporting from one hour per client to three minutes per client.

Our theory is simple: when you deliver faster, your clients are happier. They ignore the cold pitches from other agencies, they stay with you for years, and you make more money over time.

If you’re not sure where to start with AI but you want it in your agency, talk to me. Show me the workflow your team spends the most time on, and I’ll tell you honestly whether an AI delivery system is right for your agency. If it isn’t, you’ll hear that too.

Talk to Vai (co-founder)

And if you’d rather build it yourself, subscribe to my newsletter. I share at least one agent like this every week, for free.

Next step

Want a workflow like this set up for your team?

We set up AI delivery systems for marketing agencies, so each person on your team can handle more accounts.

Talk to Vai (co-founder)
Talk to Vai (co-founder)

Before you go

Not ready to talk? Get your next client deliverable out faster anyway.

Take the 230 AI skills I give agencies for free. They work in Claude, ChatGPT and Gemini, in 10 folders, one for each service you sell.

Use your agency email. I'll send the link there.