Browserless MCP¶

Official MCP server from Browserless — the leading headless Chrome SaaS used by Apify, n8n, and thousands of operators. AI agents describe what they want in natural language and Browserless executes it on production-grade Chrome infrastructure.

What It Does¶

Navigate: Browse any website with full JavaScript rendering
Scrape: Extract structured data from any page (tables, text, prices, listings)
Screenshot: Capture full-page or element-specific screenshots
PDF Generation: Convert any web page to PDF
Form Interaction: Fill forms, click buttons, submit data
Performance: Get Core Web Vitals, Lighthouse scores, and page speed metrics

Why It Matters for Operators¶

Browserless is already the production standard for headless Chrome at scale. This MCP removes the final barrier — instead of writing Playwright scripts:

Agent: "Go to competitor.com/pricing, find their enterprise plan price, and tell me if it changed since last week"
→ Browserless navigates → extracts pricing table → compares to stored value → reports delta

No more page.goto(), page.waitForSelector(), or CSS selector debugging. The AI agent handles the intent; Browserless handles the browser.

Setup¶

Prerequisites¶

Browserless account (browserless.io)
API token (Dashboard → API Keys)
Or self-host Browserless (open source, Docker image)

Claude Desktop¶

{
  "mcpServers": {
    "browserless": {
      "type": "streamable-http",
      "url": "https://chrome.browserless.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_BROWSERLESS_TOKEN"
      }
    }
  }
}

Self-Hosted¶

{
  "mcpServers": {
    "browserless": {
      "type": "streamable-http",
      "url": "http://localhost:3000/mcp",
      "headers": {}
    }
  }
}

Hermes Agent¶

mcp_servers:
  browserless:
    type: streamable-http
    url: "https://chrome.browserless.io/mcp"
    headers:
      Authorization: "Bearer ${BROWSERLESS_TOKEN}"

Tools¶

Tool	Description
`navigate`	Go to a URL with full JS rendering
`screenshot`	Capture page or element screenshots
`scrape`	Extract text, tables, and structured data
`click`	Click any element by text, selector, or position
`type`	Type into input fields
`pdf`	Generate PDF from any URL
`performance`	Get Core Web Vitals and Lighthouse scores
`execute_script`	Run arbitrary JavaScript on the page
`wait_for`	Wait for element, text, or network idle
`extract_links`	Get all links on a page
`get_network_requests`	Monitor XHR/fetch calls

Pricing¶

Browserless Cloud: - Free: 1,000 sessions/month, 1 concurrent browser - Starter: $49/month — 10,000 sessions, 5 concurrent - Growth: $199/month — 50,000 sessions, 20 concurrent - Business: $499/month — 200,000 sessions, 50 concurrent - Enterprise: Custom — unlimited sessions, dedicated infrastructure

Self-hosted: Free (open source, Docker: docker run -p 3000:3000 browserless/chrome)

Use Cases¶

Competitive Price Monitoring¶

Agent: "Every Monday at 8 AM, scrape the pricing pages of our top 5 competitors and alert me if anything changed"
→ navigate to each → extract pricing → diff against baseline → alert

Agent: "Check our top 20 landing pages for Core Web Vitals issues and list the 10 slowest"
→ navigate → performance for each → sort by LCP → report

Form Automation¶

Agent: "Register us for this industry conference — my email is [x], use the early bird code"
→ navigate to registration → type form fields → click submit → screenshot confirmation

Data Extraction at Scale¶

Agent: "From this list of 100 company URLs, extract the CEO name and contact email from each"
→ batch_navigate → scrape contact pages → extract email patterns → return CSV

Visual Regression Testing¶

Agent: "Take screenshots of our checkout flow on mobile and desktop and compare to last week's baseline"
→ navigate to checkout → screenshot(mobile) → screenshot(desktop) → diff against baseline

Self-Hosting¶

Browserless is open source. For operators who want full control:

docker run -d \
  -p 3000:3000 \
  -e "CONCURRENT=10" \
  -e "TOKEN=your-secret-token" \
  -e "MAX_QUEUE_LENGTH=100" \
  browserless/chrome:latest

Then point the MCP at http://localhost:3000/mcp.

Self-hosted benefits: - No session limits (bounded only by your hardware) - No third-party data exposure - Custom Chrome flags and extensions - VPN/geo-targeted browsing

Limitations¶

JavaScript-heavy sites: Some SPAs may require wait_for or explicit delays. The MCP handles most cases but extremely dynamic pages may need scripted approaches.
CAPTCHA: Browserless does not solve CAPTCHAs. For sites with CAPTCHA gates, use the LinkedIn MCP's managed infrastructure or similar solutions.
Session limits: Free tier is generous (1,000 sessions) but high-frequency scraping requires a paid plan.
Concurrency: Cloud plans have concurrency caps — large batch jobs queue automatically.