Bright Data Web MCP

Instructions

Implementation Reference

• browser_tools.js:206-214 (handler)

  The execute function implements the tool logic: gets the current browser page and takes a screenshot (full page optional), returns image content.

  execute: async({full_page = false})=>{
      const page = await (await require_browser()).get_page();
      try {
          const buffer = await page.screenshot({fullPage: full_page});
          return image_content({buffer});
      } catch(e){
          throw new UserError(`Error taking screenshot: ${e}`);
      }
  },

• browser_tools.js:199-205 (schema)

  Zod schema defining the input parameters: optional full_page boolean.

  parameters: z.object({
      full_page: z.boolean().optional().describe([
          'Whether to screenshot the full page (default: false)',
          'You should avoid fullscreen if it\'s not important, since the '
          +'images can be quite large',
      ].join('\n')),
  }),

• browser_tools.js:307-320 (registration)

  The tool object is included in the exported 'tools' array for registration (conditionally if API_TOKEN is set).

  export const tools = process.env.API_TOKEN ? [
      scraping_browser_navigate,
      scraping_browser_go_back,
      scraping_browser_go_forward,
      scraping_browser_links,
      scraping_browser_click,
      scraping_browser_type,
      scraping_browser_wait_for,
      scraping_browser_screenshot,
      scraping_browser_get_text,
      scraping_browser_get_html,
      scraping_browser_scroll,
      scraping_browser_scroll_to,
  ] : [scraping_browser_activation_instructions];