How chrome-mcp differs from Playwright MCP, Puppeteer MCP and Browser Use, and when you would pick each.

Every browser MCP server answers the same question differently: whose browser does the agent drive? That one decision decides whether you can read a page that needs a login, whether your 2FA has to be redone, and what happens when the agent goes somewhere it should not.

The short version

	chrome-mcp	Playwright MCP	Puppeteer MCP	Browser Use
Browser	Your running Chrome, via an MV3 extension	Playwright-launched browser by default; an extension bridge mode exists	Puppeteer-launched Chromium	Its own Chromium (Playwright/CDP)
Logged-in sessions	Yes, whatever is already open	Only with a persistent profile or the bridge mode	Only with a persistent profile	Only with a persistent profile or cookies you pass in
Default policy	Deny-all: no domains, no mutations, no eval, no downloads until you allow them	Open	Open	Open
Multi-tab fan-out	`batch` tool, parallel or serial, per-tab serialisation	One page at a time per session	One page at a time	Agent-managed
Page reads	Text, markdown, HTML, accessibility snapshot with refs, snapshot diff	Accessibility snapshot, screenshots	Screenshots, eval	Vision plus DOM extraction
Secret handling	Password values always blanked; opt-in `--redact` for tokens and keys	None	None	None
Audit trail	`history.jsonl` per task: URL, verdict, duration, bytes, secrets scrubbed	None	None	Agent logs
Install	`npx` plus load-unpacked extension	`npx`	`npx`	`pip install`
Runs as	MCP server (stdio)	MCP server	MCP server	Python library and agent
Licence	MIT	Apache-2.0	MIT	MIT

Column details are accurate to the best of our knowledge at the time of writing; the other projects move fast, so check their READMEs for the current state.

Pick chrome-mcp when

The page you need is behind a login you have already done in Chrome: a dashboard, an admin panel, a CRM, a store back office.
You do not want credentials, cookies or session tokens anywhere near a config file.
The agent should be able to read broadly but only act on the domains you name.
You need to drive many tabs at once, with real concurrency and no cross-talk.
You want an answer to "what did the agent do in my browser" after the fact.

Pick Playwright MCP when

You are writing or debugging end-to-end tests and want the same browser the test suite uses.
The target is public, or you are happy to script the login.
You need Firefox or WebKit.

Pick Puppeteer MCP when

You want the smallest possible reference implementation to fork.

Pick Browser Use when

You want a full autonomous agent in Python rather than tools for the agent you already have.
Vision-based navigation matters more than a structured tool surface.

What chrome-mcp does not do

It does not launch a browser. The extension is required; without it no tool can run.
It does not support Firefox, Safari or WebKit.
It does not run headless. It drives a real Chrome window on a machine you own.
It does not bypass Chrome's own security model. What the extension cannot reach (other extensions' pages, chrome:// pages), the agent cannot either.

Compare