Chrome DevTools MCP vs Playwright MCP
The short answer: Chrome DevTools MCP is for debugging a page: network requests, console errors, performance. Playwright MCP is for driving a page: click, fill, move on. Both start from nothing in every new session. If your agent drives the same pages again and again, there is a third option, Dejo, which remembers the route. We measured all three.
Side by side
| Chrome DevTools MCP | Playwright MCP | Dejo | |
|---|---|---|---|
| made by | Google's Chrome team | Microsoft | open source (MIT) |
| best at | debugging a live page | driving a page step by step | repeating website tasks |
| browsers | Chrome | Chrome, Edge, Firefox, WebKit | Chrome or Chromium |
| network, console, performance | yes, in depth | some | console errors and failed requests |
| remembers a task between sessions | no | no | yes |
| tool calls, 3 tasks × 10 sessions (our test) | 149 | 199 | 96 |
| tokens, same test | 7.1M | 8.7M | 4.6M |
| stays signed in | yes, its own persistent profile | yes, a persistent profile | yes, named profiles side by side |
| needs | Node | Node | Python 3.11+ |
| license | Apache 2.0 | Apache 2.0 | MIT |
Versions in our test: chrome-devtools-mcp@1.8.0, @playwright/mcp@0.0.80. Product facts from each project's official docs, checked 18 October 2026.
Driving vs debugging
Steve Kinney put it well in his article on driving vs debugging the browser: Playwright drives, DevTools debugs.
Chrome DevTools MCP gives your agent what you would open DevTools for. Its README says it lets your coding agent "control and inspect a live Chrome browser". Use it when a page is slow, a request fails, or something breaks in the console, and you want the agent to find out why.
Playwright MCP gives your agent a browser to work through: open, click, type, read. Every action returns a snapshot of the page so the model can decide the next step. Use it to get something done on a page, or to explore a flow while writing a test.
The two overlap. Both can click and both can read a page. In our test they did the same multi-step tasks, and both answered every one of their 30 sessions.
What we measured
Three tasks, each done ten times, every time in a brand-new Claude session (Claude Sonnet 5, medium effort):
| 10 sessions each | Chrome DevTools MCP | Playwright MCP | Dejo |
|---|---|---|---|
| GitHub: open Issues, count the open ones | 33 calls · 1.7M tokens | 83 calls · 3.4M | 28 calls · 1.4M |
| an author's page: read the birth date | 60 calls · 2.6M | 60 calls · 2.6M | 27 calls · 1.3M |
| a book: read its price | 56 calls · 2.8M | 56 calls · 2.8M | 41 calls · 1.9M |
| all three | 149 · 7.1M | 199 · 8.7M | 96 · 4.6M |
Chrome DevTools MCP beat Playwright MCP clearly on the GitHub task and was about the same on the other two. Neither got cheaper from one session to the next: run 10 cost about what run 1 did.
Dejo cost the most on run 1, because the agent explores and saves the route. From then on, each session was two tool calls. That is where the difference comes from. Full method and raw data: the benchmark.
Which one should you use?
- "Why is this page broken?" Chrome DevTools MCP.
- "Write a test for this flow." Playwright MCP, or Microsoft's Playwright CLI.
- "Check this dashboard every morning" / "download this month's invoice". Dejo. The first run teaches it; every run after is one call.
- "Do this once, on one page." Any of them. On one page with a one-line answer, Dejo saves nothing, because there is nothing to remember.
They are separate MCP servers, so you can install more than one. Dejo's server tells your agent to call dejo_run first for website tasks; you can still ask for the others by name.
Setting each one up
Chrome DevTools MCP and Playwright MCP run through npx; see their READMEs: Chrome DevTools MCP, Playwright MCP.
Dejo, for Claude Code:
pip install dejo-mcp
playwright install chromium
claude mcp add --scope user dejo -- dejo-mcp
Other agents: Codex, Cursor. Looking at more options? See Playwright MCP alternatives.
FAQ
What is Chrome DevTools MCP?
An MCP server from Google's Chrome team that lets a coding agent control and inspect a live Chrome browser: pages, the console, network requests and performance. It is free and open source under the Apache 2.0 license.
Is Chrome DevTools MCP better than Playwright MCP?
For debugging a page, usually yes: it gives the agent the network, console and performance data a developer would open DevTools for. For driving a page through steps, Playwright MCP is built for that. In our test, Chrome DevTools MCP used fewer calls and tokens than Playwright MCP on one of three tasks and about the same on the other two.
Does Chrome DevTools MCP keep me signed in?
Yes. By default it reuses a Chrome profile it keeps in its own folder, so sign-ins persist between runs. Its isolated option uses a temporary profile instead.
Can I use Chrome DevTools MCP and Dejo together?
Yes. A good split: Dejo for the website tasks you repeat, Chrome DevTools MCP for finding out why a page is broken.