scraperapi-mcp-server

# ScraperAPI MCP Server: What It Is, How to Set It Up in Claude, Cursor & Windsurf — Plus Every Plan Compared, Credit Costs Explained, and Whether It's Actually Worth It You know that moment when you're using Claude or Cursor and you think, *"I wish this thing could just go grab the data from that website for me right now"*? That's exactly the itch the ScraperAPI MCP server scratches. Instead of bouncing between your AI assistant and a browser, copying and pasting like it's 2015, you just type a prompt — and your LLM goes and fetches the web data itself. This article walks through the whole thing: what the ScraperAPI MCP server actually is, why it matters, how to get it running in under five minutes, and whether the underlying ScraperAPI plans are worth the money. We'll also break down every pricing tier with the real numbers — not just the headline credits that can be a bit misleading. --- **What Is the ScraperAPI MCP Server, and Why Should You Care?** MCP stands for Model Context Protocol — an open standard originally introduced by Anthropic that lets AI assistants connect to external tools and data sources in a standardized way. Think of it as a USB-C standard for AI integrations. Before MCP, getting your LLM to talk to a web scraping service meant writing custom integration code, managing HTTP calls yourself, and generally dealing with a lot of glue logic. MCP removes all of that. The ScraperAPI MCP server is ScraperAPI's official implementation of this protocol. Once configured, it sits between your LLM client (Claude Desktop, Cursor, Windsurf, Cline, VS Code + GitHub Copilot — the list keeps growing) and ScraperAPI's infrastructure. When you tell Claude "scrape this product page and give me the price history," Claude routes that request through the MCP server, ScraperAPI handles the actual scraping — including JavaScript rendering, proxy rotation, CAPTCHA bypassing, and anti-bot evasion — and the clean result comes right back into the conversation. The practical upside is significant. ScraperAPI maintains a pool of over 40 million IPs across 50+ countries, and their infrastructure handles the messy stuff that makes DIY web scraping a nightmare: rotating proxies, solving CAPTCHAs, rendering JavaScript-heavy pages, and dealing with bot-detection systems like Cloudflare, Turnstile, DataDome, and PerimeterX. Your AI agent gets to focus on the thinking part; ScraperAPI handles the fetching part. --- **Two Modes: Remote vs. Local** The ScraperAPI MCP server ships in two flavors, and which one you want depends on how much setup friction you're comfortable with. **Remote (Hosted) Mode** connects your LLM client directly to ScraperAPI's hosted endpoint at `https://mcp.scraperapi.com/mcp`. There's nothing to install locally. You just add a URL and your API key to your client's configuration file, and you're live. This is the easiest option and the one most people should start with. **Local (Self-Hosted) Mode** runs the MCP server as a process on your own machine, launched automatically by your LLM client when it needs it. You can install it via Python (`pip install scraperapi-mcp-server`) or run it via Docker. The local mode gives you slightly more control over things like request timeouts and rate limiting, but for most use cases, the remote mode works just as well. --- **What Tools Does It Expose?** Once the MCP server is connected, your AI assistant gets access to a whole suite of tools — far more than just "go fetch this URL." Here's a breakdown of what's available as of the March 2026 expansion: - **`scrape`** — The core tool. Retrieves any web page, bypassing anti-bot systems. Supports JavaScript rendering (`render=true`), geo-targeting (`country_code`), premium and ultra-premium proxies, mobile/desktop user agent switching, and output in Markdown, plain text, CSV, or JSON. - **Structured Data Endpoints (SDEs)** — Pre-parsed, structured JSON data from major platforms without needing to wrangle raw HTML: - `google_search` — Parsed Google SERP results - `google_news` — Google News results - `google_jobs` — Google Jobs listings - `google_shopping` — Google Shopping results - `google_maps_search` — Google Maps business/place results - `amazon_product` / `amazon_search` / `amazon_offers` — Amazon product and marketplace data - `walmart_search` / `walmart_product` / `walmart_category` / `walmart_review` - `ebay_search` / `ebay_product` - `redfin_for_sale` / `redfin_for_rent` / `redfin_search` / `redfin_agent` - **Crawler tools** (`crawler_job_start`, `crawler_job_status`, `crawler_job_delete`) — Full async web crawling through MCP, not just single-page fetches. - **AI Parser tools** (`ai_parser_create`, `ai_parser_parse_url`, etc.) — Build reusable parsers from example URLs and apply them at scale. The smart-agent context feature (the "MCP Skill," installable via `npx skills add scraperapi/scraperapi-skills`) helps the AI automatically choose the right tool for each task — so it knows to use `google_search` when you ask for SERP data versus `scrape` when you want raw page content. --- **Setup Guide: Getting the ScraperAPI MCP Server Running** Let's do the actual setup. First, [sign up for a ScraperAPI account](https://www.scraperapi.com/?fp_ref=coupons) (there's a free tier with 1,000 credits/month to start) and grab your API key from the dashboard. **Option A: Remote Server (Recommended for Most Users)** For Claude Desktop, open the settings → Developer tab → Edit Config, and paste: json { "mcpServers": { "ScraperAPI": { "type": "http", "url": "https://mcp.scraperapi.com/mcp", "headers": { "Authorization": "Bearer YOUR_SCRAPERAPI_API_KEY" } } } } That's it. Restart Claude Desktop, and the ScraperAPI tools will appear in your tool panel. **Option B: Local Server via Python** Install the package: bash pip install scraperapi-mcp-server Then add this to your client's MCP config file: json { "mcpServers": { "ScraperAPI": { "command": "python", "args": ["-m", "scraperapi_mcp_server"], "env": { "API_KEY": "YOUR_SCRAPERAPI_API_KEY" } } } } **Option C: Local Server via Docker** json { "mcpServers": { "ScraperAPI": { "command": "docker", "args": [ "run", "-i", "-e", "API_KEY=${API_KEY}", "--rm", "scraperapi-mcp-server" ] } } } For **Cursor**: Settings → Tools & Integrations → Add MCP Server → Manual → paste the JSON config. For **Windsurf**: Settings → Cascade settings → MCP server section → gear icon → paste into `mcp_config.json`. For **Cline (VS Code extension)**: Click the MCP Servers icon → Configure tab → "Configure MCP Servers" → paste into `cline_mcp_settings.json`. For **Claude Code** (CLI): bash claude mcp add scraperapi -e API_KEY=YOUR_SCRAPERAPI_API_KEY -- python -m scraperapi_mcp_server Once it's running, you can keep things as simple as typing: *"Scrape this URL and extract the product price"* — or get more specific: *"Scrape this page with JS rendering enabled and geo-targeted to Germany, then return the results as JSON."* --- **Understanding Credits: The Part That Trips Most People Up** Before we get to the plans, let's talk about how ScraperAPI's credit system actually works — because the headline numbers on the pricing page can be misleading if you read them as a flat request count. Every plan comes with a certain number of API credits per month. But a "credit" isn't a one-to-one exchange for a page. Here's the multiplier table: | Request Type | Credits Per Request | |---|---| | Standard (no extra params) | 1 | | `render=true` (JS rendering) | 10 | | `premium=true` (premium proxies) | 10 | | `screenshot=true` | 10 | | `premium=true` + `render=true` | 25 | | `ultra_premium=true` | 30 | | `ultra_premium=true` + `render=true` | 75 | | Cloudflare / DataDome / PerimeterX bypass | 10 | On top of that, certain high-value domains have fixed multipliers: | Domain | Credits Per Request | |---|---| | Amazon | 5 | | Google / Bing SERP | 25 | | LinkedIn | 30 | The practical implication: if you're on the Hobby plan (100,000 credits) and you're scraping JS-rendered pages, you're actually getting around 10,000 rendered scrapes — not 100,000. If you're scraping Google SERPs on the Business plan (3,000,000 credits), the real capacity is about 120,000 SERP queries. Neither of those numbers is bad — they're just different from what you might assume on first glance. ScraperAPI does provide a `sa-credit-cost` response header in every API call so you always know what a request actually cost, and there's a `urlcost` API endpoint where you can look up the credit cost for any URL before running a large job. --- **All ScraperAPI Plans Compared** 👉 [Start for free or explore all plans](https://www.scraperapi.com/pricing/?fp_ref=coupons) | Plan | Monthly Price | API Credits | Concurrent Threads | Geotargeting | Overage | Best For | |---|---|---|---|---|---|---| | Free | $0 | 1,000 / mo | 5 | US & EU | No | Testing, proof-of-concept | | Hobby | $49 / mo | 100,000 | 20 | US & EU | No | Small personal projects | | Startup | $149 / mo | 1,000,000 | 50 | US & EU | No | Low-volume workflows | | Business | $299 / mo | 3,000,000 | 100 | Global (country-level) | No | Production-grade, moderate scale | | Scaling | $475 / mo | 5,000,000 | 200 | Global | Yes (PAYG) | Growing ops, most popular | | Professional | $975 / mo | 10,500,000 | 300 | Global | Yes (PAYG) | High-volume recurring jobs | | Advanced | $1,975 / mo | 21,500,000 | 500 | Global | Yes (PAYG) | Continuous multi-source pipelines | | Enterprise | Custom | 22,000,000+ | 500+ | Global | Yes | Custom data ops, dedicated support | All plans include: JS rendering, premium proxies, CAPTCHA/anti-bot bypass, rotating proxy pools, and unlimited bandwidth. Annual billing knocks 10% off every paid tier. A few things worth noting: - The **Free tier** gives you 1,000 credits per month with 5 concurrent connections — genuinely enough to evaluate the MCP server and run small experiments, no credit card required. - The **Scaling plan** is listed as "most popular" and is the first tier where pay-as-you-go overage kicks in. If you're serious about using the MCP server in production, this is the tier where you stop worrying about hitting the wall mid-month. - **Business** is the first tier with global country-level geotargeting and unlimited analytics retention, which matters if your scraping jobs span multiple regions. - The **Hobby and Startup plans** don't support PAYG overage — you'll need to upgrade when credits run out. 👉 [Get started with the Hobby plan ($49/mo)](https://www.scraperapi.com/pricing/?fp_ref=coupons) 👉 [Jump to the Scaling plan ($475/mo)](https://www.scraperapi.com/pricing/?fp_ref=coupons) --- **What the MCP Server Is Actually Good At** If you're using an AI coding assistant like Cursor or Windsurf for any of the following workflows, the ScraperAPI MCP server changes the game: **Competitive pricing research.** Instead of manually checking 20 competitor product pages, you tell your AI agent to pull and compare them. With the Amazon SDE, you can pull structured product data across ASINs without writing a single line of scraping code. **SERP monitoring.** Asking Claude to check your keyword rankings against competitor pages used to require a whole separate tool. With `google_search` available as an MCP tool, it's a one-liner prompt. **Real estate and market data.** The Redfin integration (added as part of the March 2026 expansion) lets agents pull live listing data, rental prices, and agent profiles directly into AI workflows. **Content summarization from live sources.** Instead of pasting article text into Claude manually, you just give it a URL and ask for a summary. The `scrape` tool handles the fetching; Claude handles the thinking. **E-commerce monitoring at scale.** Walmart, eBay, and Amazon SDEs return pre-parsed JSON — meaning your AI doesn't have to fight with raw HTML at all. --- **A Few Things to Keep in Mind** No tool is perfect, and ScraperAPI has some real trade-offs worth knowing before you commit. The **credit multiplier system** is powerful but requires attention. Running ultra-premium + JS rendering at 75 credits per request on a 100,000-credit Hobby plan gives you just over 1,300 requests per month. That's fine for testing; it's not fine for production anti-bot workflows on a tight budget. **Overage handling is tier-gated.** If you're on Hobby, Startup, or Business and you burn through your credits before month end, you're stuck until you upgrade or roll over. This isn't a dealbreaker, but it means entry-tier plans require a bit more upfront planning about expected usage. **The free tier is genuinely small.** 1,000 credits per month with 5 concurrent connections is useful for experimenting with the MCP server setup, but you'll burn through it fast if you start doing rendered scrapes. --- **Real-World Prompts to Try Once You're Set Up** Once your MCP server is connected, here are some prompts that showcase what it can do: > *"Scrape the homepage of [competitor website]. If you get a 500 error, add geo-targeting for the US. If it still fails, use premium proxies."* > *"Search Google for 'best standing desks under $500' and return the top 10 results as a structured list."* > *"Pull the Amazon product page for ASIN B08N5WRWNW and extract the price, rating, and number of reviews."* > *"Start a crawl of [website] following only links that contain '/blog/', up to depth 2, and return all page titles and URLs."* The MCP Skill (the intelligent agent layer installable with `npx skills add scraperapi/scraperapi-skills`) helps your AI pick the right underlying tool automatically, so you don't have to remember which endpoint does what. --- **Is the ScraperAPI MCP Server Worth It?** If you're already using an MCP-compatible AI assistant (Claude, Cursor, Windsurf, Cline) and you do any kind of web research, data collection, or competitive analysis, this is a pretty easy yes. The setup takes about five minutes, the free tier is enough to test it meaningfully, and the integration is clean enough that it genuinely changes how fast you can move on research-heavy tasks. For solo developers or small teams doing light scraping, the Hobby plan at $49/month gives you 100,000 credits with the MCP server fully enabled — that's roughly 10,000 JS-rendered pages per month, which covers a lot of ground. For anyone running automated data pipelines, competitive monitoring, or AI agent workflows at scale, the Scaling or Professional tier is where the math starts to make a lot of sense. The fact that ScraperAPI has been around since 2018, has 10,000+ companies on the platform, and has been expanding its MCP surface continuously (the March 2026 update alone added SDEs for five major platforms plus full crawler integration) suggests this isn't a side project. It's the company's bet on how AI agents are going to interact with the web. 👉 [Create a free ScraperAPI account and get 1,000 credits today](https://www.scraperapi.com/?fp_ref=coupons)v