alterlab

AlterLab MCP Server

Give Claude, Cursor, and Windsurf the ability to scrape any website, extract structured data, and take screenshots — with automatic anti-bot bypass.

Get Started Free →   $1 free balance on signup — up to 5,000 scrapes.

---

One-Line Install

Claude Code

claude mcp add alterlab -- npx -y alterlab-mcp-server@latest

Then set your API key: export ALTERLAB_API_KEY=sk_live_... or add it to .claude.json (see full setup below).

Cursor

# Add to .cursor/mcp.json — see full config below

Smithery

npx -y @smithery/cli install alterlab-mcp-server --client claude

---

Why AlterLab Instead of WebFetch or Browser MCP?

Claude's built-in WebFetch tool and open-source browser MCP servers fail on most real-world websites. They cannot bypass Cloudflare, render JavaScript SPAs, or extract structured data.

AlterLab replaces broken fetch tools with one MCP server that actually works:

Capability	WebFetch / fetch()	Browser MCP	AlterLab MCP
Anti-bot bypass (Cloudflare, DataDome, Akamai)	No	Partial	Yes — automatic
JavaScript rendering (React, Angular, Vue SPAs)	No	Yes (slow)	Yes — headless Chromium
Structured data extraction (JSON, Schema.org)	No	No	Yes — built-in profiles
Smart tier escalation (cheapest method first)	N/A	N/A	Yes — saves 60-80%
Residential proxy rotation (195+ countries)	No	No	Yes
Screenshot and PDF capture	No	Screenshot only	Yes — both
OCR text extraction from images	No	No	Yes
Cost per request	Free (but fails)	Free (but slow)	From $0.0002

How Does AlterLab Web Scraping Work?

AlterLab uses a multi-tier scraping architecture. It automatically selects the cheapest method capable of fetching each URL:

1. Curl ($0.0002/req) — Direct HTTP for static pages, RSS feeds, public APIs
2. HTTP ($0.0003/req) — TLS fingerprint rotation for moderately protected sites
3. Stealth ($0.002/req) — Browser impersonation for Cloudflare/DataDome-protected sites
4. Light JS ($0.0025/req) — Lightweight JS extraction from server-rendered HTML
5. Browser ($0.004/req) — Full headless Chromium for JavaScript-heavy SPAs

Auto mode starts at Tier 1 and escalates only when blocked. Most websites resolve at Tiers 1-2, so $1 gets you 1,000 to 5,000 scrapes depending on the sites you target.

---

Installation

Install in Claude Desktop / Claude Code

Add to your Claude config file (~/.claude.json for Claude Code, or Settings for Claude Desktop):

{
  "mcpServers": {
    "alterlab": {
      "command": "npx",
      "args": ["-y", "alterlab-mcp-server@latest"],
      "env": {
        "ALTERLAB_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Install in Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "alterlab": {
      "command": "npx",
      "args": ["-y", "alterlab-mcp-server@latest"],
      "env": {
        "ALTERLAB_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Install in Windsurf

Add to Windsurf MCP settings (~/.codeium/windsurf/mcp_config.json):

{
  "mcpServers": {
    "alterlab": {
      "command": "npx",
      "args": ["-y", "alterlab-mcp-server@latest"],
      "env": {
        "ALTERLAB_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Install via Smithery

npx -y @smithery/cli install alterlab-mcp-server --client claude

Get Your API Key

1. Sign up free — $1 free balance on signup
2. Go to Dashboard → API Keys and copy your key
3. Paste it into the ALTERLAB_API_KEY field in your MCP config

---

Tools

alterlab_scrape — Scrape Any Webpage

Scrape a URL and return its content as markdown, text, HTML, or JSON. Automatically handles anti-bot protection with tier escalation. Returns markdown by default — optimized for LLM context windows.

"Scrape https://www.amazon.com/dp/B0BSHF7WHW and summarize the product"

Parameter	Type	Default	Description
url	string	required	URL to scrape
mode	auto | html | js | pdf | ocr	auto	Scraping mode
method / body / content_type / headers	enum / string / enum / object	GET	POST support (1.5x cost) and custom target headers
formats	(text|json|json_v2|html|markdown|rag|content|raw)[]	["markdown"]	Output formats
extraction_schema / extraction_prompt	object / string	—	LLM extraction (+$0.001)
extraction_profile / extraction_template	enum	—	auto, product, article, job_posting, faq, recipe, event, ecommerce_homepage, directory_listing (template is a deprecated alias)
extraction_provider / extraction_model	enum / string	—	BYOK LLM selection
evidence, promote_schema_org, filter_content	boolean	—	Extraction behaviour
cache, cache_ttl, force_refresh	boolean / number / boolean	—	Response caching (cache_ttl 60-86400)
render_js	boolean | "auto"	API default	Headless browser (forces Browser tier, $0.004)
use_proxy, proxy_country	boolean, string	—	Premium proxy (+$0.0002) and geo-targeting
use_own_proxy, use_system_proxy, proxy_integration_id	boolean, boolean, string	—	Bring-your-own-proxy selection
skip_tls_verification	boolean	—	Omit (recommended) = verify, fall back on invalid certs; true skip, false require valid cert
wait_for, wait_until, wait_condition	string	—	Selector / load-event waits
actions	object[] (max 20)	—	Browser actions: wait, wait_for, click, fill, press, scroll, infinite_scroll, screenshot, evaluate
screenshot, generate_pdf, ocr	boolean	—	Add-ons (+$0.0002 / +$0.0004 / +$0.001)
scroll_to_load, scroll_count, enable_scroll	boolean / 1-10 / boolean	—	Lazy-load handling
include_iframes, flatten_shadow_dom, block_images, remove_cookie_banners	boolean	—	Rendering options
section_filter	object	—	Filter for json_v2 sections
session_id, cookies, session_headers	UUID, object, object	—	Authenticated scraping
sticky_session, sticky_session_ttl	string, 300-7200	—	Reuse cookies and exit IP across calls (paid plans)
location	{country, language}	—	Geo-targeting
cost_controls	object	—	force_tier, max_tier, max_credits, prefer_cost, prefer_speed, fail_fast, time_budget (prefer_cost, prefer_speed, fail_fast also accepted at top level)
embeddings	object	—	enabled, model, chunk_size, chunk_overlap (+$0.001/page)
sync_policy	auto | wait_until_deadline	auto	Wait behaviour for long scrapes
timeout, max_response_bytes, include_raw_html	number, number, boolean	90, 5 MB, false	Limits and raw output
estimate_first	boolean	false	Run a free cost estimate first

Not exposed: source (dashboard tracking tag), sync (the tool is always synchronous), webhook_url (use alterlab_batch/alterlab_crawl webhooks). If a scrape outlives the ~85s synchronous window the tool returns a job_id; fetch it with alterlab_get_job.

alterlab_extract — Extract Structured Data

Extract structured fields from HTML, text or markdown content using pre-built profiles or custom JSON Schema. Returns clean JSON — ready for databases, spreadsheets, or downstream processing.

"Extract the product name, price, and rating from this Amazon page"

Parameter	Type	Default	Description
content	string (max 500,000)	required	HTML, text or markdown you already have (this tool does not fetch a URL)
content_type	html | text | markdown	html	Type of content
extraction_profile / extraction_template	enum	—	Profile: auto, product, article, job_posting, faq, recipe, event, ecommerce_homepage, directory_listing
extraction_schema	object	—	Custom JSON Schema for structured output
extraction_prompt	string	—	Natural language extraction instructions
extraction_provider / extraction_model	enum / string	—	BYOK LLM selection
formats	enum[]	["json"]	Output formats
source_url, evidence	string, boolean	—	Context URL and provenance
cache, cache_ttl	auto | skip | only, number	auto	Extraction result cache

To scrape and extract in one step, use alterlab_scrape with extraction_schema.

Extraction profiles:

• Product — name, price, currency, rating, reviews, availability, images, description
• Article — title, author, published date, body text, featured image
• Job Posting — title, company, location, salary, description, requirements
• FAQ — question-answer pairs
• Recipe — ingredients, instructions, prep time, servings
• Event — name, date, location, description, organizer

alterlab_screenshot — Screenshot Any Page

Take a full-page screenshot of any URL. Returns a PNG image directly in the conversation — no URLs to copy, no files to download.

"Take a screenshot of our landing page at https://alterlab.io"

Parameter	Type	Default	Description
url	string	required	URL to screenshot
wait_for	string	—	CSS selector to wait for before capture
wait_until	enum	networkidle	networkidle, domcontentloaded, or load
timeout	number	90	Timeout in seconds (1-300)
use_proxy, proxy_country	boolean, string	—	Premium proxy and geo-targeting
session_id, cookies	UUID, object	—	Screenshot pages behind a login
location	{country, language}	—	Geo-targeting

alterlab_estimate_cost — Estimate Before You Scrape

Check how much a scrape will cost before running it. Returns the predicted tier, cost per request, and confidence level.

"How much would it cost to scrape linkedin.com?"

Parameter	Type	Default	Description
url	string	required	URL to estimate
mode	enum	auto	Scraping mode
method	GET | POST	GET	POST costs 1.5x
render_js	boolean | "auto"	false	Include JS rendering cost
use_proxy, use_own_proxy	boolean	false	Include proxy cost / BYOP discount
screenshot, generate_pdf, ocr	boolean	—	Include add-on cost

alterlab_get_job — Fetch an Async Job

Fetch the status and result of a scrape that returned a job_id (a scrape that outlived the synchronous wait window). Parameter: job_id (string, required).

alterlab_check_balance — Check Your Credits

Check your account balance, total deposited, and total spent. No parameters needed.

"Check my AlterLab balance"

alterlab_list_sessions — List Stored Sessions

List all stored sessions for authenticated scraping. Sessions contain cookies for specific domains, allowing you to scrape content behind login walls.

"List my stored sessions"

alterlab_create_session — Create a Session

Create a new stored session with cookies from a logged-in browser. The session is stored securely and can be reused across multiple scrape requests.

"Create an Amazon session with these cookies: session-id=abc123, session-token=xyz789"

Parameter	Type	Default	Description
name	string	required	Human-readable name (e.g., "My Amazon Account")
domain	string	required	Domain (e.g., "amazon.com")
consent	boolean	required	Must be true: explicit GDPR consent to store the cookies (the API rejects creation otherwise)
cookies	Record<string,string> | cookie objects[]	—	Cookies as key-value pairs or {name, value, domain, path, ...} objects
headers	Record<string,string>	—	Custom headers (e.g. a User-Agent)
expires_at, notes	string	—	Expiry (ISO 8601) and notes
auto_refresh_enabled, notify_before_expiry, notify_on_failure	boolean	true	Health checks and notifications
organization_id, shared	UUID, boolean	—	Organization sharing

alterlab_validate_session — Validate a Session

Check whether a stored session is still active and its cookies are valid.

"Is my Amazon session still valid?"

Parameter	Type	Default	Description
session_id	string (UUID)	required	Session ID to validate

alterlab_delete_session — Delete a Session

Permanently delete a stored session and its cookies.

"Delete session abc-123-def"

Parameter	Type	Default	Description
session_id	string (UUID)	required	Session ID to delete

---

Authenticated Scraping

AlterLab MCP supports scraping pages that require authentication. This enables AI agents to access user-specific content like order histories, account dashboards, and member-only pricing.

How It Works

1. Create a session with cookies from a logged-in browser using alterlab_create_session
2. Scrape authenticated pages by passing the session_id to alterlab_scrape
3. Manage sessions with list, validate, and delete tools

Example: Check Amazon Prime Pricing

User: "What's my Prime member price for this product?"

Claude: [calls alterlab_list_sessions → finds Amazon session]
Claude: [calls alterlab_scrape with session_id for authenticated pricing]
Claude: "The Prime member price is $24.99 (public price: $34.99)"

Inline Cookies vs Stored Sessions

• Stored sessions (session_id): Best for repeated access to the same domain. Create once, reuse across requests.
• Inline cookies (cookies): Best for one-off authenticated requests where you don't need to save the session.

---

What Can You Do with AlterLab MCP?

Research and Analysis

Ask Claude to scrape and analyze websites in real-time:

• "Scrape the top 5 results from this Google search and summarize them"
• "Extract all product prices from this Amazon category page"
• "Compare the pricing pages of these 3 competitors"

Code Generation with Real Data

Let Cursor or Windsurf fetch live data while building:

• "Scrape this API documentation page and generate TypeScript types from it"
• "Extract the color palette from this website and create a Tailwind config"
• "Screenshot this design and recreate it in React"

Content and SEO

Use Claude to analyze content at scale:

• "Scrape this blog post and suggest improvements for SEO"
• "Extract all FAQ entries from this help center and create a structured dataset"
• "Compare our landing page to the competitor's and identify gaps"

Monitoring and Alerts

Build agentic workflows that watch the web:

• "Check if this product is back in stock"
• "Scrape this page daily and alert me when the price drops below $50"
• "Monitor this job board for new senior engineering positions"

---

Pricing — Pay-As-You-Go Web Scraping

No subscriptions. No monthly minimums. Add balance and use it whenever you need it.

Base Scraping Costs

Tier	Method	Cost per Request	Use Case
Curl	Direct HTTP	$0.0002	Static pages, RSS feeds, public APIs
HTTP	TLS fingerprinting	$0.0003	Sites with basic bot detection
Stealth	Browser impersonation	$0.002	Cloudflare, DataDome, PerimeterX protected sites
Light JS	JSON extraction	$0.0025	Server-rendered pages needing structured data
Browser	Headless Chromium	$0.004	Full JavaScript SPAs (React, Angular, Vue)

Optional Add-Ons

Add-On	Extra Cost	Description
JavaScript Rendering	forces Browser tier	No separate add-on: render_js bills the Browser tier ($0.004)
PDF Generation	+$0.0004	PDF of the rendered page
LLM Extraction / Embeddings	+$0.001	extraction_schema / extraction_prompt, embeddings
POST requests	x1.5	method: "POST" costs 1.5x the tier price
Own proxy (BYOP)	-20%	use_own_proxy discount on the tier price
Search	$0.001 / $0.004	duckduckgo / google and bing engines
Screenshot Capture	+$0.0002	Full-page PNG screenshot
Premium Proxy	+$0.0002	Geo-targeted residential proxy (195+ countries)
OCR Text Extraction	+$0.001	Extract text from images on the page

$1 = 5,000 light scrapes. New accounts get $1 free balance on signup.

---

Environment Variables

Variable	Required	Default	Description
ALTERLAB_API_KEY	Yes	—	Your API key (get one free)
ALTERLAB_API_URL	No	https://api.alterlab.io	API base URL (for self-hosted or development)

---

Frequently Asked Questions

How do I add web scraping to Claude, Cursor, or Windsurf?

Install the AlterLab MCP server. For Claude Code, run claude mcp add alterlab -- npx -y alterlab-mcp-server@latest and set your ALTERLAB_API_KEY. For Claude Desktop, Cursor, or Windsurf, add the JSON config block to your MCP settings file. Once configured, your AI assistant can scrape any URL, extract structured data, and take screenshots directly in conversation.

Can Claude scrape websites that are behind Cloudflare or anti-bot protection?

Yes. AlterLab automatically handles Cloudflare, DataDome, PerimeterX, Akamai, and other anti-bot systems. It uses a multi-tier approach that starts with the cheapest method and escalates only when blocked. You don't need to configure anything — anti-bot bypass is fully automatic.

What is an MCP server and how does it work with Claude?

MCP (Model Context Protocol) is Anthropic's open standard for connecting AI assistants to external tools and data sources. An MCP server is a small program that exposes tools — like web scraping — that Claude, Cursor, or Windsurf can call during a conversation. The AlterLab MCP server gives your AI assistant 23 tools: scrape, extract, screenshot, crawl, search, map, batch scraping, session management, beta features, and more.

How is AlterLab different from Firecrawl, ScrapingBee, or Apify MCP servers?

AlterLab starts at $0.0002 per request — 5-20x cheaper than most scraping APIs — because it only uses expensive browser rendering when a site actually requires it. Smart tier escalation means you pay for what each site needs, not the maximum. AlterLab also includes built-in structured data extraction with pre-built profiles (product, article, job posting, etc.).

Can I scrape Amazon, Walmart, and other e-commerce sites from Claude?

Yes. AlterLab handles all major e-commerce anti-bot protection. Use the alterlab_extract tool with extraction_profile: "product" to get structured JSON: product name, price, currency, rating, review count, availability, and images — ready for analysis, comparison, or data pipelines.

Can Cursor scrape documentation and generate code from it?

Yes. With AlterLab MCP installed in Cursor, you can ask it to scrape API documentation, library docs, or any reference page and generate TypeScript types, API clients, or component code from the live content. This is more reliable than relying on the LLM's training data, which may be outdated.

Does AlterLab MCP work with JavaScript-heavy sites (React, Angular, Vue)?

Yes. Use render_js: true or set mode: "js" to enable full headless Chromium rendering. AlterLab renders the complete page including all JavaScript, waits for dynamic content to load, then extracts content from the fully rendered DOM. This works for React, Angular, Vue, Next.js, and any other JavaScript framework.

What output format is best for AI and LLM context windows?

Use markdown (the default). It preserves document structure — headings, tables, lists, links — while being 60-80% smaller than raw HTML. Claude, GPT-4, and other LLMs process markdown significantly better than HTML. AlterLab's markdown output is specifically optimized for LLM context windows.

Is there rate limiting?

Free-tier accounts have rate limits. Adding any balance removes rate limits. The MCP server includes automatic retry with exponential backoff for transient rate limit errors (429).

Can I use this MCP server for large-scale scraping?

Yes. For parallel scraping, use alterlab_batch to scrape up to 100 URLs in a single request — much faster than sequential calls. For crawling entire domains, use alterlab_crawl. You can also build agentic workflows that scrape many URLs sequentially, or use the n8n integration for workflow automation.

---

Error Handling

The MCP server returns helpful error messages with suggested next actions:

Error	What Happens	Suggested Action
401 Unauthorized	Invalid API key	Check ALTERLAB_API_KEY is set correctly
402 Insufficient Credits	Balance too low	Run alterlab_check_balance, add funds
403 Forbidden	Site blocked the request	Try render_js: true + use_proxy: true
429 Rate Limited	Too many requests	Automatic retry with backoff
504 Gateway Timeout	Scrape took too long	Increase timeout, simplify request

---

Agentic Workflows — AI Agents That Scrape the Web

AlterLab MCP turns any AI agent into a web-capable agent. Instead of relying on stale training data, your agent can fetch live information from any website during execution.

Use Cases for AI Agents

• Research agents — scrape multiple sources, cross-reference facts, generate reports with citations
• Data pipeline agents — extract structured product/pricing/job data on a schedule
• Competitive intelligence — monitor competitor pages, track pricing changes, detect new features
• Content generation — scrape real data to ground LLM output in facts, not hallucinations
• Lead enrichment — scrape company websites to enrich CRM records with live data
• Compliance monitoring — check regulatory pages, terms of service, and policy updates

Compatible AI Frameworks and Clients

Works with any tool that supports MCP (Model Context Protocol):

• Claude Desktop / Claude Code — Anthropic's AI assistant
• Cursor — AI-powered code editor
• Windsurf — Codeium's AI IDE
• Cline — VS Code AI assistant
• GitHub Copilot — via MCP configuration
• Custom agents — any framework using the MCP SDK (Python, TypeScript, Go)
• n8n — workflow automation via n8n-nodes-alterlab

---

Contributing

git clone https://github.com/RapierCraft/alterlab-mcp-server.git
cd alterlab-mcp-server
npm install
npm run build

Support

• API Documentation
• Dashboard & API Keys
• GitHub Issues
• [email protected]

License

MIT