Crawlbase review

Pricing: the sliding ladder

As of October 2026, Crawlbase's pricing page sets the pay-as-you-go rate in bands of monthly successful requests. Like tax brackets, each band's rate applies only to the requests inside it. You are billed in arrears for what you used, "so there is nothing to expire".

Beyond that the rate drops to $0.05, $0.04 and finally $0.02 per 1,000. Worked through: 100,000 requests cost $3 + $20 + (89 × $0.60 = $53.40) = $76.40, or $0.76 per 1,000; 25,000 cost $31.40 ($1.26 per 1,000); 10,000 cost $21 ($2.10 per 1,000). Small jobs are expensive per page; the ladder rewards volume.

Subscriptions run from Starter at $99 a month (200,000 credits, $0.50 per 1,000) through $199 (500K) and $349 (1M) to Growth at $599 (2M, $0.30 per 1,000); usage past the block continues "at the bracket your volume already reached". On plain requests, 200,000 on pay-as-you-go cost $127.50 against Starter's $99, and the break-even sits near 143,000 a month. Below that, stay on pay-as-you-go. Model your own volumes in the cost calculator.

JavaScript and difficulty multipliers

Rendering in a real browser "doubles the credits of any request, whatever the site tier". Because the doubled credits also move you down the ladder faster, 10,000 rendered requests count as 20,000: $3 + $20 + (9 × $0.60) = $28.40, or $2.84 per 1,000. On top of that, each site falls into a difficulty tier, applied automatically: Standard ×1, Moderate ×1.5, Complex ×2, Complex I ×3, Complex II ×4.5, Extreme ×9, Extreme I ×13.5 and Extreme II ×20. The pricing page says only "a small set of heavily defended sites" carries a higher one and doesn't list which domains.

That is where the cheap ladder stops being cheap. A rendered request to an Extreme II site counts as 40 credits: in the $0.60 band, 1,000 of them cost about $24, against a flat $1.50 at Bright Data's Web Unlocker or $16.08 for Zyte's hardest browser tier. The saving grace is that "a request that fails or is blocked costs nothing", so the sites Crawlbase can't open cost you time, not money.

The MCP server

Crawlbase's MCP server runs locally through npx and is documented for Claude Desktop, Cursor, Windsurf and n8n. The configuration from its GitHub README:

{
  "mcpServers": {
    "crawlbase": {
      "type": "stdio",
      "command": "npx",
      "args": ["@crawlbase/mcp@latest"],
      "env": {
        "CRAWLBASE_TOKEN": "your_token_here",
        "CRAWLBASE_JS_TOKEN": "your_js_token_here"
      }
    }
  }
}

The normal token handles plain HTML requests; the JS token handles rendered pages and screenshots. Three tools are exposed: crawl (HTML), crawl_markdown (clean Markdown) and crawl_screenshot (a full-page screenshot). There is no search tool: an agent can read a page you point it at but cannot find pages, so pair it with a search API such as ScrapingBee or PacketStream. Being local, your tokens sit in a config file on each machine; ZenRows, ScrapingBee and Firecrawl offer hosted servers instead. At agent volumes the ladder works well: 1,000 fetches a month cost $3.00 and 25,000 cost $31.40, with no plan.

Independent evidence

The only independent result we have is Proxyway's Web Scraping API Report 2025 (tested October 2025, published 3 December 2025; Proxyway uses affiliate links). Crawlbase came seventh of ten at 56.72% across 15 protected sites, with 0% on five of them and a slow 23.94-second average response. Yet it scored 96.91% on G2, where Oxylabs managed 0% and ScrapingBee 5%. Results swing by target, so spend the free credits on your own URLs before trusting any average.

Where it fits

Our guides pick products by use case. Crawlbase is picked in:

• Search & fetch APIs for AI agents guide: Pay-as-you-go fetching with JavaScript rendering

The case is the billing model: no subscription, only successful requests charged, and JavaScript rendering, with HTML, Markdown or screenshots back. Other tools fit better in these cases:

• Your agent needs web search. Crawlbase has no search tool. Bright Data's SERP API charges $1.50 per 1,000 successful requests pay-as-you-go and PacketStream $2.50 per 1,000 from a prepaid balance; ScrapingBee bundles Google search and JavaScript fetch in a hosted MCP server from $19 a month.
• You want a free monthly allowance or a hosted server. ZenRows gives 5,000 free credits a month with a hosted MCP server; Firecrawl gives 1,000 and adds LLM-ready Markdown and crawling.
• Your targets are heavily protected. Zyte API reached 93.14% in the same 2025 test, and Bright Data's Web Unlocker charges a flat $1.50 per 1,000 with no difficulty multipliers.

See the AI-agent guide and the scraping API guide for the full comparison; how picks are chosen is explained on the methodology page.

Pros and cons