Chrome DevTools MCP vs Browser Tools MCP

When to pick Chrome DevTools MCP

• Choose Chrome DevTools MCP when the agent needs to reproduce a browser workflow: open a page, click controls, fill forms, and inspect the result. Its Puppeteer-based automation waits for action results, which fits investigations that require repeated interactions.
• Pick it when a slow page needs trace-level investigation. Use performance_start_trace and performance_stop_trace to record activity, then performance_analyze_insight to examine performance findings alongside console messages and network requests.
• Use it when memory behavior is part of the bug. take_heapsnapshot gives the agent a memory investigation tool alongside evaluate_script, screenshots, and page snapshots.
• Choose it when you want the server to launch Chrome for the task or attach to a running instance through a remote debugging port. An extension-based connection is not required for this workflow.

When to pick Browser Tools MCP

• Choose Browser Tools MCP when the bug is visible in your existing logged-in Chrome session. Its DevTools extension captures that session's console output and network activity without requiring you to recreate the login in a separate test profile.
• Pick it when you want to select an element yourself and ask the agent to inspect it. getSelectedElement reads the selection in the Elements panel; getPageInfo identifies the page being inspected.
• Use it when debugging several tabs with DevTools open. listBrowserTabs returns identifiers for explicit targeting, and telemetry retention is separate for each tab, so a noisy page does not displace another page's history.
• Choose it for a focused log review or audit workflow. Filter console and network results, clear captured logs with wipeLogs before reproducing a bug, or request accessibility, performance, SEO, and best-practices audits through dedicated tools.

Main differences

• Browser connection differs. Chrome DevTools MCP launches or remotely attaches to Chrome. Browser Tools MCP runs a connector inside its MCP process and pairs with a Chrome extension. Network capture begins when DevTools opens; reload to capture requests that completed earlier.
• Interaction tools differ. Chrome DevTools MCP provides click and fill_form through its automation workflow. Browser Tools MCP also supports navigation and soft input, but its softClick and softType actions are untrusted browser events, require input to be enabled, and can report that a real gesture is needed.
• Performance investigation differs. Chrome DevTools MCP combines traces, performance insights, heap snapshots, and Lighthouse audits. Browser Tools MCP exposes separate Lighthouse audit tools. Its audits launch a separate browser, so audit execution should not be confused with extension capture from your logged-in tab.
• Authentication differs. Chrome DevTools MCP requires no server authentication in the supplied configuration. Browser Tools MCP protects its loopback HTTP API and extension WebSocket with a per-run bearer token; the extension obtains it through a pairing route restricted to browser-extension origins.
• Data handling differs. Chrome DevTools MCP exposes browser content to the MCP client, collects usage statistics by default, and may send trace URLs to Google CrUX. Disable those collections with --no-usage-statistics and --no-performance-crux. Browser Tools MCP scrubs captured credentials, disables header capture by default, and withholds storage values unless requested; its connector and extension connection stay on loopback.

Setup

For Chrome DevTools MCP, configure your MCP client to run npx -y chrome-devtools-mcp@latest. Install Node.js LTS, npm, and current stable Chrome or newer. The browser starts when a tool first needs it, rather than when the MCP client merely connects.

For Browser Tools MCP, configure npx -y @agentdeskai/browser-tools-mcp@latest with Node 22.19 or newer. Load the repository's chrome-extension directory through Load unpacked at chrome://extensions, with Developer mode enabled. Open DevTools on the target page to begin capture. The connector starts within the MCP server; use npx @agentdeskai/browser-tools-mcp --doctor to diagnose setup problems.