Chrome Browser
by hangwin
Chrome MCP Server turns your everyday Chrome browser — logins, cookies, and history included — into a controllable tool for AI assistants via a Chrome extension.
About
Chrome MCP Server (mcp-chrome) is a Chrome extension paired with a local Node bridge that exposes your actual running browser to MCP clients, rather than spinning up a separate automated instance the way Playwright-style tools do. Because it drives your real Chrome profile, an agent inherits your existing logins, cookies, and session state instead of needing to re-authenticate everywhere.
It ships over 20 tools spanning navigation and tab control, screenshots, network capture (both a lightweight webRequest listener and a full Debugger-API mode that includes response bodies), form filling and clicking via a unified computer-style interaction tool, bookmark and history search, and an accessibility-tree reader for structured page understanding. A standout feature is search_tabs_content, which uses a built-in vector database with WebAssembly SIMD-accelerated embeddings to run semantic search across the content of your open tabs — finding the right tab by meaning, not just by title match.
Key features
- Drives your existing Chrome browser and profile — no separate automated browser instance
- 20+ tools covering navigation, screenshots, forms, and clicks
- Two network-capture modes: webRequest listener and full Debugger API with response bodies
- Semantic, vector-based search across open tabs' content (search_tabs_content)
- Accessibility-tree page reading (chrome_read_page) with stable element identifiers
- Runs fully local for privacy — no browser data leaves your machine
Use cases
- Have an agent find and switch to the right open tab by describing its content instead of its title
- Capture network requests and response bodies while reproducing a bug on a logged-in site
- Fill out and submit a form on a site where you're already authenticated, without exporting cookies
- Search your browsing history or bookmarks by keyword as part of a research task
Available tools
chrome_navigate
Navigates to a URL with optional viewport control.
chrome_screenshot
Takes a screenshot of the page or a specific element with configurable options.
search_tabs_content
Runs AI-powered semantic search across the content of currently open browser tabs.
chrome_network_debugger_start
Starts capturing network traffic via the Chrome Debugger API, including response bodies.
chrome_read_page
Builds an accessibility tree of the current page with stable identifiers for semantic element discovery.
chrome_bookmark_add
Adds a new bookmark, with folder support.
Frequently asked questions
Does this launch a separate, isolated browser like Playwright does?
No — it controls your actual Chrome browser through an extension, so it uses your real profile, logins, and settings instead of a fresh automated instance.
Does my browsing data get sent anywhere?
No — the MCP server and its bridge run entirely on your machine, and the project describes itself as a fully local server for that reason.