Scraper Builder MCP

Build, save, and run web scrapers with AI assistance. An MCP server that lets AI agents explore websites and generate reusable TypeScript scrapers.

Why This MCP?

Other browser MCPs help AI extract data once. This one saves reusable TypeScript scrapers you can run anytime — in scripts, cron jobs, or pipelines.

Usage Examples

Creating a Scraper

Just describe what you want and provide a JSON schema:

"Use scraper-builder-mcp to create a scraper for https://example.com/jobs
Extract data matching this schema:

{ "title": "string", "company": "string", "location": "string", "url": "string" }

Save it as example_jobs and run it to verify it works"

The AI will:

1. Navigate to the page and take a screenshot
2. Explore the DOM to find the right selectors
3. Write a TypeScript scraper that outputs your schema
4. Save it to your project folder
5. CRITICAL: Run it immediately to verify it works (and fix if needed)

Tips:

• If your IDE has built-in browser tools, say "use scraper-builder-mcp" explicitly
• Scrapers save to scrapers/ and data to data/ in the current project's root directory
• Always ask the AI to run the scraper after saving to catch issues early

Running a Saved Scraper

"Run the example_jobs scraper"

Or:

"List my scrapers"

More Examples

You say...	AI does...
"Scrape the top 10 products from this page"	Explores, extracts, returns data
"Save this as a reusable scraper called products"	Saves to scrapers/products.ts
"Run the products scraper"	Runs scraper, saves data to data/products/, returns results
"The website changed, update the products scraper"	Edits the saved scraper

Quick Start

Install via npx (Recommended)

Add to your MCP config (Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "scraper-builder-mcp": {
      "command": "npx",
      "args": ["-y", "scraper-builder-mcp"]
    }
  }
}

Auto-Approve Permissions (Recommended)

To avoid confirming every tool execution (like taking screenshots or scrolling), follow these steps for your client:

Client	How to Auto-Approve
Cursor	Add scraper-builder-mcp:* to the MCP Allowlist in Settings.
Claude Desktop	Click "Always approve" when the first tool permission prompt appears.
Windsurf / Others	Approve when prompted; most clients will remember your choice for the session or permanently.

Full list of tools for granular Cursor whitelisting

If you prefer not to use a wildcard, add these individually:

scraper-builder-mcp:navigate
scraper-builder-mcp:screenshot
scraper-builder-mcp:get_page_content
scraper-builder-mcp:find_elements
scraper-builder-mcp:extract_text
scraper-builder-mcp:click
scraper-builder-mcp:type
scraper-builder-mcp:scroll
scraper-builder-mcp:wait_for
scraper-builder-mcp:execute_javascript
scraper-builder-mcp:download_resource
scraper-builder-mcp:save_scraper
scraper-builder-mcp:list_scrapers
scraper-builder-mcp:run_scraper

To show the browser window while scraping, add:

{
  "mcpServers": {
    "scraper-builder-mcp": {
      "command": "npx",
      "args": ["-y", "scraper-builder-mcp"],
      "env": { "HEADED": "true" }
    }
  }
}

Chromium will be installed automatically on first use.

Install from Source

git clone https://github.com/emilneander/scraper-builder-mcp.git
cd scraper-builder-mcp
npm install
npm run build

Then point your MCP config to dist/index.js.

Tools

Scraper Workflow

Tool	Description
save_scraper	Save a reusable TypeScript scraper script
list_scrapers	List all saved scrapers
run_scraper	Execute a saved scraper and get fresh data

Browser Automation

Tool	Description
navigate	Go to a URL for web scraping
screenshot	Capture page for visual analysis
get_page_content	Get HTML or simplified DOM for finding selectors
find_elements	Test CSS selectors and verify matches
extract_text	Preview data extraction results
click	Click on an element
type	Type into input fields
scroll	Scroll the page (handles lazy loading)
wait_for	Wait for elements or network idle
execute_javascript	Run JS in browser context
download_resource	Download files with browser cookies/auth

License

MIT