# agent-browser-mcp [Health: Active]

**Category:** 📂 Browser Automation  
**Repository:** https://github.com/minhlucvan/agent-browser-mcp  
**GitHub Stars:** 14  
**npm Downloads (last month):** 3139  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/agent-browser-mcp

## Description
MCP server integrating with Vercel's agent-browser for AI-driven browser automation

## Tools
Capabilities this server exposes over MCP:

- **browser_navigate** — Navigate to a URL
- **browser_go_back** — Navigate back in browser history
- **browser_go_forward** — Navigate forward in browser history
- **browser_reload** — Reload the current page
- **browser_click** — Click on an element using selector or accessibility locator
- **browser_fill** — Fill a text input field with a value
- **browser_type** — Type text character by character (triggers key events)
- **browser_hover** — Hover over an element
- **browser_scroll** — Scroll the page or a specific element
- **browser_select** — Select an option from a dropdown
- **browser_check** — Check a checkbox or radio button
- **browser_uncheck** — Uncheck a checkbox
- **browser_press** — Press a keyboard key (Enter, Escape, Tab, etc.)
- **browser_get_text** — Get text content from an element or the entire page
- **browser_get_html** — Get HTML content (inner or outer)
- **browser_get_attribute** — Get an attribute value from an element
- **browser_get_url** — Get the current page URL
- **browser_get_title** — Get the current page title
- **browser_snapshot** — Get accessibility tree snapshot for AI-friendly element references
- **browser_is_visible** — Check if an element is visible
- **browser_is_enabled** — Check if an element is enabled
- **browser_is_checked** — Check if a checkbox or radio button is checked
- **browser_screenshot** — Take a screenshot of the page or a specific element
- **browser_pdf** — Generate a PDF of the current page
- **browser_new_session** — Create a new isolated browser session with optional viewport
- **browser_close_session** — Close a browser session
- **browser_wait_for_selector** — Wait for an element to appear (attached, detached, visible, hidden)
- **browser_wait_for_navigation** — Wait for navigation to complete
- **browser_get_cookies** — Get cookies, optionally filtered by URLs
- **browser_set_cookies** — Set cookies with full options (domain, path, expiry, etc.)
- **browser_clear_cookies** — Clear all cookies
- **browser_evaluate** — Execute JavaScript code in the browser context
- **browser_get_console** — Get console messages from the browser
- **browser_get_network** — Get network requests made by the browser

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "agent-browser-mcp": {
    "command": "npx",
    "args": ["agent-browser-mcp"]
  }
}
```

## Documentation

## What agent-browser-mcp does

The agent-browser-mcp MCP server connects an MCP-compatible client to Vercel's agent-browser browser automation CLI. It gives an AI agent tools for opening pages, navigating browser history, reloading content, interacting with controls, reading page data, and inspecting browser state.

The tool set covers common browser workflows: clicking, filling and typing into inputs, selecting dropdown values, checking controls, hovering, scrolling, and pressing keys. Agents can retrieve page text, HTML, URLs, titles, element attributes, and accessibility snapshots. They can also capture screenshots, create PDFs, execute JavaScript in the browser context, and inspect console messages and network requests.

Use agent-browser-mcp when an agent needs to operate a website through browser-visible interactions rather than only call an HTTP API. The repository lists web scraping, AI-assisted testing, multi-step forms, screenshot analysis, authenticated workflows, bulk data entry, and page monitoring as use cases.

## How it works

The MCP process delegates browser operations to agent-browser. Install the CLI and its Chromium browser separately, then start the MCP server with `npx agent-browser-mcp` or install the npm package and run it from your client configuration.

Selectors can use semantic patterns such as roles, accessible names, visible text, ARIA attributes, placeholders, and test IDs. Standard CSS selectors are also supported. Accessibility snapshots provide structured references that an agent can use to locate elements without relying exclusively on fragile CSS or XPath expressions.

Each browser session can be created independently and may use its own viewport. Sessions isolate cookies, storage, and navigation history, allowing parallel or separate workflows. Session-specific operations receive a session ID, and the session can be closed when the workflow finishes.

## Setup and configuration

Node.js 18 or newer and the `agent-browser` CLI are prerequisites. Install the CLI globally with `npm install -g agent-browser`, then run `agent-browser install` to download Chromium. Linux installations may also need `agent-browser install --with-deps` for system dependencies.

Start the server directly with `npx agent-browser-mcp`, or install it with `npm install agent-browser-mcp`. The README includes stdio configuration examples for Claude Desktop, VS Code, and Cursor, plus a `claude mcp add agent-browser -- npx agent-browser-mcp` command for Claude Code.

If the CLI is not available on the system PATH, set `AGENT_BROWSER_PATH` to its location in the MCP client's environment configuration. Windows users should account for documented upstream issues with native PowerShell and CMD shells; the README recommends WSL until those issues are resolved.

## Tools and capabilities

Available operations include:

- Navigation with `browser_navigate`, history controls, and reload.
- Interaction through click, fill, type, hover, scroll, select, check, uncheck, and key-press tools.
- Extraction of text, HTML, attributes, URLs, titles, and accessibility snapshots.
- Visibility, enabled-state, and checked-state inspection.
- Screenshots, PDFs, JavaScript evaluation, console retrieval, and network inspection.
- Cookie retrieval, cookie creation, and cookie clearing.
- Session creation with an optional viewport and session closure.
- Waiting for selectors or navigation to complete.

The underlying agent-browser implementation is described as using a fast Rust CLI with a Node.js fallback, structured output, accessibility-oriented snapshots, and streaming output.

## Limitations and notes

The MCP server depends on the separately installed agent-browser CLI and Chromium browser. It does not provide a browser without those prerequisites. The custom path setting is available when the CLI cannot be found through PATH.

Windows native shells have known upstream issues according to the README, so WSL may be needed. Browser actions also depend on selectors matching the target page; semantic locators, accessibility properties, text, data attributes, and CSS selectors are supported, but the page must expose a usable target.

The documented server configuration uses stdio through a command such as `npx agent-browser-mcp`. No environment variables are listed as mandatory for a normal installation.

_Full upstream README: https://allmcps.com/mcp/agent-browser-mcp/readme_