Concurrent Browser MCP

Instructions

Implementation Reference

• src/tools.ts:1001-1189 (handler)

  The primary handler function for 'browser_get_markdown'. It retrieves the browser instance and uses page.evaluate to run a comprehensive JavaScript function that traverses the DOM and converts HTML elements to Markdown format, handling headings, links, lists, tables, etc. Supports options like includeLinks, maxLength, and selector.
  private async getMarkdown(instanceId: string, options: { includeLinks: boolean; maxLength: number; selector?: string; }): Promise<ToolResult> { const instance = this.browserManager.getInstance(instanceId); if (!instance) { return { success: false, error: `Instance ${instanceId} not found` }; } try { // JavaScript function to extract page content and convert to Markdown const markdownContent = await instance.page.evaluate((opts) => { const { includeLinks, maxLength, selector } = opts; // Select the root element to process const rootElement = selector ? document.querySelector(selector) : document.body; if (!rootElement) { return 'Specified element or page content not found'; } // HTML to Markdown conversion function function htmlToMarkdown(element: any, depth = 0) { let markdown = ''; const indent = ' '.repeat(depth); for (const node of element.childNodes) { if (node.nodeType === Node.TEXT_NODE) { const text = node.textContent?.trim(); if (text) { markdown += text + ' '; } } else if (node.nodeType === Node.ELEMENT_NODE) { const el = node as Element; const tagName = el.tagName.toLowerCase(); switch (tagName) { case 'h1': markdown += `\n\n# ${el.textContent?.trim()}\n\n`; break; case 'h2': markdown += `\n\n## ${el.textContent?.trim()}\n\n`; break; case 'h3': markdown += `\n\n### ${el.textContent?.trim()}\n\n`; break; case 'h4': markdown += `\n\n#### ${el.textContent?.trim()}\n\n`; break; case 'h5': markdown += `\n\n##### ${el.textContent?.trim()}\n\n`; break; case 'h6': markdown += `\n\n###### ${el.textContent?.trim()}\n\n`; break; case 'p': const pText = htmlToMarkdown(el, depth); if (pText.trim()) { markdown += `\n\n${pText.trim()}\n`; } break; case 'br': markdown += '\n'; break; case 'strong': case 'b': markdown += `**${el.textContent?.trim()}**`; break; case 'em': case 'i': markdown += `*${el.textContent?.trim()}*`; break; case 'code': markdown += `\`${el.textContent?.trim()}\``; break; case 'pre': markdown += `\n\`\`\`\n${el.textContent?.trim()}\n\`\`\`\n`; break; case 'a': const href = el.getAttribute('href'); const linkText = el.textContent?.trim(); if (includeLinks && href && linkText) { if (href.startsWith('http')) { markdown += `[${linkText}](${href})`; } else { markdown += linkText; } } else { markdown += linkText || ''; } break; case 'ul': case 'ol': markdown += '\n'; const listItems = el.querySelectorAll('li'); listItems.forEach((li, index) => { const bullet = tagName === 'ul' ? '-' : `${index + 1}.`; markdown += `${indent}${bullet} ${li.textContent?.trim()}\n`; }); markdown += '\n'; break; case 'blockquote': const quoteText = el.textContent?.trim(); if (quoteText) { markdown += `\n> ${quoteText}\n\n`; } break; case 'div': case 'section': case 'article': case 'main': // Recursively process container elements markdown += htmlToMarkdown(el, depth); break; case 'table': // Simplified table processing const rows = el.querySelectorAll('tr'); if (rows.length > 0) { markdown += '\n\n'; rows.forEach((row, rowIndex) => { const cells = row.querySelectorAll('td, th'); const cellTexts = Array.from(cells).map(cell => cell.textContent?.trim() || ''); markdown += '| ' + cellTexts.join(' | ') + ' |\n'; if (rowIndex === 0) { markdown += '|' + ' --- |'.repeat(cells.length) + '\n'; } }); markdown += '\n'; } break; case 'script': case 'style': case 'nav': case 'footer': case 'aside': // Ignore these elements break; default: // For other elements, continue recursive processing of child elements markdown += htmlToMarkdown(el, depth); break; } } } return markdown; } // Extract page title const title = document.title; const url = window.location.href; // Generate Markdown content let content = `# ${title}\n\n**URL:** ${url}\n\n`; content += htmlToMarkdown(rootElement); // Clean up extra line breaks and spaces content = content .replace(/\n{3,}/g, '\n\n') .replace(/[ \t]+/g, ' ') .trim(); // Truncate content if exceeds maximum length if (content.length > maxLength) { content = content.substring(0, maxLength) + '\n\n[Content truncated...]'; } return content; }, options); return { success: true, data: { markdown: markdownContent, length: markdownContent.length, truncated: markdownContent.length >= options.maxLength, url: instance.page.url(), title: await instance.page.title() }, instanceId }; } catch (error) { return { success: false, error: `Get markdown failed: ${error instanceof Error ? error.message : error}`, instanceId }; } }
• src/tools.ts:464-486 (schema)

  Defines the input schema for the browser_get_markdown tool, specifying parameters like instanceId (required), includeLinks, maxLength, and optional selector.
  inputSchema: { type: 'object', properties: { instanceId: { type: 'string', description: 'Instance ID' }, includeLinks: { type: 'boolean', description: 'Whether to include links', default: true }, maxLength: { type: 'number', description: 'Maximum content length in characters', default: 10000 }, selector: { type: 'string', description: 'Optional CSS selector to extract content from specific element only' } }, required: ['instanceId']
• src/tools.ts:461-488 (registration)

  Registers the browser_get_markdown tool in the getTools() method by including it in the returned Tool[] array.
  { name: 'browser_get_markdown', description: 'Get page content in Markdown format, optimized for large language models', inputSchema: { type: 'object', properties: { instanceId: { type: 'string', description: 'Instance ID' }, includeLinks: { type: 'boolean', description: 'Whether to include links', default: true }, maxLength: { type: 'number', description: 'Maximum content length in characters', default: 10000 }, selector: { type: 'string', description: 'Optional CSS selector to extract content from specific element only' } }, required: ['instanceId'] } }
• src/tools.ts:578-584 (registration)

  In the executeTools switch statement, routes calls to 'browser_get_markdown' to the getMarkdown handler method.
  case 'browser_get_markdown': return await this.getMarkdown(args.instanceId, { includeLinks: args.includeLinks ?? true, maxLength: args.maxLength || 10000, selector: args.selector });