Browserless MCP

What browserless/browserless-mcp does

The browserless/browserless-mcp MCP server exposes Browserless.io capabilities to MCP clients. It is intended for agents that need to retrieve content from modern websites, search the web, inspect site structure, automate browser interactions, or review Browserless activity.

The server covers both direct data collection and interactive workflows. Single-page scraping can return Markdown or HTML, while mapping and crawling support broader site discovery. Browser sessions can be driven through snapshots and ref-based actions, including clicks, typing, scrolling, evaluation, screenshots, multiple tabs, captcha handling, and file transfers.

How it works

The client sends MCP tool calls to Browserless through the hosted endpoint. Browserless executes the browser or API operation and returns the resulting content, metadata, metrics, session information, or account data to the client.

For interactive pages, browserless_agent uses a repeated snapshot, planning, action, and re-snapshot process. browserless_skill supplies recipes for page mechanics such as shadow DOM, cookie notices, modal dialogs, captchas, dynamic content, screenshots, and tabs. Authentication profiles can be listed and then passed to other tools to reuse saved cookies and origins.

Setup and configuration

The browserless/browserless-mcp MCP server is configured with a Browserless API token. Create or obtain the token from Browserless.io, then add the hosted MCP URL to the client configuration:

{
  "mcpServers": {
    "browserless": {
      "url": "https://mcp.browserless.io/mcp?token=your-token-here"
    }
  }
}

The documented setup runs remotely and does not require a local installation. The README names Claude Desktop, Cursor, VS Code, and Windsurf as supported MCP clients. A free Browserless tier is available, according to the supplied material.

Tools and capabilities

• browserless_smartscraper extracts one webpage as Markdown or HTML, including JavaScript-heavy pages and pages requiring anti-bot handling.
• browserless_search performs web, news, or image searches, with optional result scraping, geographic targeting, and time filters.
• browserless_map discovers URLs through sitemaps and link extraction, optionally returning titles and descriptions.
• browserless_crawl scrapes discovered pages with controls for depth, path filters, sitemap strategy, and scrape options.
• browserless_performance runs Lighthouse audits for accessibility, best practices, performance, PWA, and SEO, with optional category filters and performance budgets.
• browserless_function runs custom Puppeteer JavaScript with a page object and optional context. The returned { data, type } controls the payload and content type.
• browserless_export retrieves a page through the Browserless export API and detects native output such as HTML, PDF, or images.
• browserless_agent operates persistent browser sessions and supports multi-tab work, screenshots, captcha solving, live URLs, and upload or download workflows.
• browserless_profiles, browserless_account, and browserless_usage expose saved authentication profiles, account details, and consumption data.
• browserless_sessions inspects active, persistent, and recorded sessions and can download a self-contained replay page. browserless_logs reports recent request outcomes, timing, stopping reasons, and cost.

Limitations and notes

The browserless/browserless-mcp MCP server depends on a Browserless API token and the hosted Browserless service. Log history availability depends on the account plan. Account inspection does not return API token values. Browserless usage and session information reflects the account associated with the current token, and authentication profiles must already be saved there before another tool can reuse them.

For multi-page content, use crawling rather than the single-page scraper. Use mapping when the goal is to enumerate URLs without scraping every page. Custom browser behavior requires Puppeteer JavaScript, while more complex interactive page mechanics may require loading a matching skill before or during agent operation.