MCP Browser Extension — Claude Code MCP server: drive your own Chrome
An MCP server that points your agent at the Chrome window you already have open, sessions and cookies included. Useful where a site has no API.
claude mcp add mcp-browser-extension -- npx -y mcp-browser-extensionWhat MCP Browser Extension does
There are two pieces here: a Chrome extension and a small MCP server started with npx (package and command mcp-browser-extension). Grab the extension from the Chrome Web Store or load it unpacked. The server listens on a local WebSocket, the extension dials in, and tabs are driven through chrome.scripting and chrome.tabs. No separate Chromium is launched, so nothing works until the extension is in place.
Nothing is reachable until you say so. With no flags the agent reads zero pages. Hosts open up one at a time — --allow-domain mail.google.com --allow-domain github.com — and *.example.com also covers subdomains. Reads are checked, not just clicks, and the extension enforces the same rule on its end. Page changes, eval, downloads and uploads each need their own opt-in. Password fields never come back in a read, and --redact strips tokens, API keys and JWTs.
The agent works in background tabs, so the page in front of you is not navigated or clicked away; one batch call can pull several tabs at once. Every call lands in an audit log with the URL, the verdict and the size of the response. Access can be cut from the popup or Settings, and the block holds after a restart. An expired session returns [AUTH_REQUIRED] rather than a vague timeout. Credentials are never stored, and the server does not sign in for you.
What it helps with
- read a Gmail thread in a tab you are already signed into
- go through pull requests and settings on a private GitHub repo
- pull numbers off a dashboard that has no API
- keep the agent inside one allowed domain
- read several open tabs in a single call
Example prompts for the agent
- Open the analytics dashboard tab and pull the revenue table for last month
- Read the latest Gmail thread and summarize what they are asking for
- Walk the open pull requests on GitHub and list the ones waiting on my review
How to install MCP Browser Extension
Platform CLI
claude mcp add mcp-browser-extension -- npx -y mcp-browser-extensionPlatform CLI
codex mcp add mcp-browser-extension -- npx -y mcp-browser-extensionConfig file
{
"mcpServers": {
"mcp-browser-extension": {
"command": "npx",
"args": [
"-y",
"mcp-browser-extension"
]
}
}
}Config file
{
"servers": {
"mcp-browser-extension": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"mcp-browser-extension"
]
}
}
}Platform CLI
gemini mcp add mcp-browser-extension npx -y mcp-browser-extensionStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about MCP Browser Extension
How do I install MCP Browser Extension in Claude Code?
Run claude mcp add mcp-browser-extension -- npx -y mcp-browser-extension, then add the extension from the Chrome Web Store. claude mcp list shows whether the server came up.
What does MCP Browser Extension do?
It gives the agent access to the Chrome you already have open, so pages load under your existing sessions and cookies with no second login. The extension and a local WebSocket server do the work.
Is any setup required before the first run?
Yes — with no flags, access stays fully closed. Name your hosts with --allow-domain, and turn on page changes, eval, downloads and uploads through their own flags.
