Playwright MCP vs Chrome DevTools MCP: Browser Agents Compared (2026)

Pick Playwright MCP when the AI needs to click through a site in Chrome, Firefox, WebKit or Edge, fill forms and check flows. Pick Chrome DevTools MCP when the AI needs to debug a page in Chrome. It records performance traces, reads network and console data, runs Lighthouse and finds memory leaks. Many web teams install both and use each for its strength.

Checked against: the Playwright MCP README and the Chrome DevTools MCP README on 28 October 2026. eSEOspace did not test either product.

Both are free, open-source servers that run on your own computer and drive a real browser. Microsoft publishes Playwright MCP. The Chrome DevTools team publishes Chrome DevTools MCP, which its README now calls "Chrome DevTools for agents."

At a glance

Where Playwright MCP is stronger

Cross-browser testing. The README's --browser option accepts "chrome, firefox, webkit, msedge." If your users are on Safari, WebKit coverage matters. Chrome DevTools MCP says other Chromium browsers "may work, but this is not guaranteed."

Structured, model-friendly input. Playwright works from the accessibility tree, "not pixel-based input." The README calls this "Deterministic tool application" that "Avoids ambiguity common with screenshot-based approaches." A browser_snapshot tool returns the page structure, which the README says "is better than screenshot."

Opt-in capability groups. Core tools cover navigation, clicks, typing, forms, tabs, dialogs, uploads and console messages. Extra groups load only when you ask with --caps, including network, storage, DevTools, PDF, coordinate-based "vision" clicks and test assertions. Fewer tools mean a smaller tool list for the model.

Testing workflows. The testing capability adds assertion tools, and a --mobile flag emulates a phone. Microsoft positions MCP for "exploratory automation, self-healing tests, or long-running autonomous workflows."

Docker and HTTP. You can run it in Docker or start it with --port for HTTP clients. The Docker image "only supports headless chromium at the moment."

Microsoft adds an honest caveat. For coding agents, it says a Playwright CLI with skills is often "more token-efficient" than MCP.

Where Chrome DevTools MCP is stronger

Performance work. Chrome DevTools MCP uses DevTools "to record traces and extract actionable performance insights." Its tool reference lists performance_start_trace, performance_stop_trace and performance_analyze_insight. For Core Web Vitals work, that is the main draw.

Deeper debugging. The debugging group includes lighthouse_audit, get_css_styles, evaluate_script, console tools and screencasts. Network tools return full request details. The README says console messages come "with source-mapped stack traces."

Memory and extensions. A memory group of 14 tools takes and analyzes heap snapshots. Extension tools install, reload and trigger Chrome extensions. Progressive Web App tools can install and launch PWAs.

Bundled skills. The package ships skills for accessibility debugging, cookie debugging, LCP (Largest Contentful Paint) and memory leaks. In Claude Code you can install the server and skills as one plugin.

Attach to real Chrome. --browserUrl connects to a running, debuggable Chrome. --autoConnect attaches to a local Chrome 144 or newer. The README also points to Android debugging in its advanced guide.

Early WebMCP support. An experimental WebMCP group lists and runs tools that a web page exposes. It is off unless you pass --categoryExperimentalWebmcp.

Pricing

Both servers are free and open source under Apache 2.0. Neither vendor sells a paid tier for the MCP server.

Your costs are indirect. Each tool call and page snapshot uses model tokens in your AI app. Microsoft's own README says verbose accessibility trees cost context, which is why it recommends the CLI for some coding agents. Chrome DevTools MCP offers --slim "for basic browser tasks only." If you run either server in the cloud, you pay for that hosting.

Security and admin controls

Neither is a security boundary. Playwright's README says so directly. It adds that --allowed-origins and --blocked-origins do "not serve as a security boundary" and do not affect redirects. Chrome DevTools MCP warns it "exposes content of the browser instance to the MCP clients allowing them to inspect, debug, and modify any data in the browser or DevTools."

Profiles and sessions. Playwright uses a persistent profile by default. Use --isolated to keep the profile in memory. Chrome DevTools MCP also supports --isolated and --headless. Avoid browser sessions logged in to admin panels, payment tools or customer data. Our guide to sandboxing local MCP servers shows safer setups.

Data sent to Google. Chrome DevTools MCP says "Data collection is enabled by default" for usage statistics such as tool success rates and latency. Turn it off with --no-usage-statistics, or set the CI environment variable. Performance tools "may send trace URLs to the Google CrUX API." Use --no-performance-crux to stop that. The Playwright README does not describe any usage data collection.

Script execution. Both can run JavaScript in the page, through browser_evaluate in Playwright and evaluate_script in Chrome DevTools. Web pages are untrusted input, so watch for prompt injection.

Network exposure. If you start Playwright with --port or in Docker on 0.0.0.0, anyone who can reach that port can drive the browser. Read exposed local MCP servers.

Which to choose

Pick Playwright MCP if:

• You need to test in Firefox, WebKit or Edge, not just Chrome.
• The job is user flows: sign-up, checkout, forms and navigation.
• You want a long-running agent that keeps browser state.

Pick Chrome DevTools MCP if:

• You are chasing slow pages, layout shifts or poor Core Web Vitals.
• You need Lighthouse, network detail, console stack traces or heap snapshots.
• Your team builds and debugs mainly in Chrome.

Use both if you want flow testing and performance debugging on the same site. Keep each scoped to a staging environment. Our browser testing playbook shows one way to split the work.

Next step

eSEOspace sets up browser MCP servers in Claude Code for website QA and performance work. See our Claude Code development service, or browse more MCP comparisons. Full profiles: Playwright MCP and Chrome DevTools MCP.

Frequently asked questions