# MCP Server Browser

> MCP Server Browser is an open-source MCP server for browser automation. MCP server giving LLM clients browser automation via structured page data. It can run locally on your machine or as a hosted remote server.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, API, self-hosted, vscode
- **Repository:** https://github.com/bytedance/ui-tars-desktop/tree/HEAD/packages/agent-infra/mcp-servers/browser
- **MCP Registry name:** io.github.bytedance/mcp-server-browser
- **Categories:** [browser automation](https://ai-powered.com/categories/browser-automation)
- **Listing updated:** 2026-11-18

## About

The README describes this as a Model Context Protocol server that lets an LLM drive a browser. Instead of working from screenshots, it reads Puppeteer's accessibility tree and labelled element index, which the README says keeps context usage low and removes the need for a vision model in the default mode.

A vision mode is available as an option for pages where structured data is not enough. The server can run over stdio for local MCP clients, or be started with a port flag to expose SSE and streamable HTTP endpoints. Viewport size, coordinate factors and user agent can be set per request through HTTP headers.

It suits developers wiring up AI clients such as Cursor, VS Code, Windsurf or Claude Desktop who want those clients to navigate pages, fill forms and extract text or links. It can attach to remote or local browsers, and the README also documents a Docker setup that combines VNC with the MCP endpoint.

## Features

- **Accessibility-based page reading:** Uses Puppeteer's labelled accessibility tree so the model works on structured element data rather than pixels.
- **Optional vision mode:** A flag enables screenshot-based interaction for layouts where structured data falls short.
- **Multiple transports:** Runs over stdio locally or as SSE and streamable HTTP servers on a chosen port.
- **Runtime browser configuration:** Viewport size, vision coordinate factors and user agent can be set per request via HTTP headers.
- **Remote and local browser support:** Can connect to a CDP or WebSocket browser endpoint, or launch a local browser such as Chrome, Edge or Firefox.
- **Browser tool set:** Exposes tools for navigation, clicking, form filling, tabs, screenshots, JavaScript evaluation and downloads.

## Good for

- Letting an AI coding assistant open and inspect a web page during development
- Automating form filling and navigation steps from an LLM agent
- Extracting page text, markdown or links for an agent to read
- Running a shared headless browser behind an MCP endpoint in Docker

## Install

### Claude Code (Terminal)

```sh
claude mcp add mcp-server-browser -- npx -y @agent-infra/mcp-server-browser
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "mcp-server-browser": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-infra/mcp-server-browser"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "mcp-server-browser": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-infra/mcp-server-browser"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "mcp-server-browser": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@agent-infra/mcp-server-browser"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "mcp-server-browser": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-infra/mcp-server-browser"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "mcp-server-browser": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-infra/mcp-server-browser"
      ]
    }
  }
}
```

## Tools

- `browser_navigate`: Navigate to a URL
- `browser_click`: Click an element on the page
- `browser_close`: Close the browser when the task is done
- `browser_close_tab`: Close the current tab
- `browser_evaluate`: Execute JavaScript in the browser console
- `browser_form_input_fill`: Fill out an input field
- `browser_get_clickable_elements`: Get the clickable or hoverable or selectable elements on the current page
- `browser_get_download_list`: Get the list of downloaded files
- `browser_get_markdown`: Get the markdown content of the current page
- `browser_get_text`: Get the text content of the current page
- `browser_go_back`: Go back to the previous page
- `browser_go_forward`: Go forward to the next page
- `browser_hover`: Hover an element on the page
- `browser_new_tab`: Open a new tab
- `browser_press_key`: Press a key on the keyboard
- `browser_read_links`: Get all links on the current page
- `browser_screenshot`: Take a screenshot of the current page or a specific element
- `browser_scroll`: Scroll the page
- `browser_select`: Select an element on the page with index
- `browser_switch_tab`: Switch to a specific tab
- `browser_tab_list`: Get the list of tabs
- `browser_vision_screen_capture`: Take a screenshot of the current page for vision mode
- `browser_vision_screen_click`: Click left mouse button on the page with vision and snapshot

## FAQ

### How do I install it?

The README gives configuration snippets that run the package with npx, and one-click install links for VS Code and Cursor.

### Does it need a vision model?

No. The README says the default mode works on structured accessibility data, and vision mode is optional.

### Which transports are supported?

Stdio by default, and SSE and streamable HTTP when started with a port argument.

### Which browsers can it use?

The README lists chrome, edge and firefox as browser options, and also mentions connecting to CDP or WebSocket endpoints.

### Is MCP Server Browser free?

Yes. MCP Server Browser is open source, so you can use and self-host it for free.

### How do I add MCP Server Browser to Claude Code?

Run this command in your terminal: claude mcp add mcp-server-browser -- npx -y @agent-infra/mcp-server-browser

### What are the best alternatives to MCP Server Browser?

The closest alternatives to MCP Server Browser are Playwright MCP, Selenium MCP, playwright_mcp, MCP Browser and Agent Browser.

## Where this comes from

- Description: written from the project's README and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Playwright MCP](https://ai-powered.com/mcp/playwright-mcp): Playwright MCP is an open-source MCP server for browser automation.
- [Selenium MCP](https://ai-powered.com/mcp/selenium-mcp-2): Selenium MCP is an open-source MCP server for browser automation.
- [playwright_mcp](https://ai-powered.com/mcp/playwright-mcp-3): playwright_mcp is an open-source MCP server for browser automation.
- [MCP Browser](https://ai-powered.com/mcp/mcp-browser-2): MCP Browser is an open-source MCP server for browser automation.
- [Agent Browser](https://ai-powered.com/mcp/agent-browser-3): Agent Browser is an open-source MCP server for browser automation.
- [MCP Browser](https://ai-powered.com/mcp/mcp-browser-3): MCP Browser is an open-source MCP server for browser automation.
- [Chrome DevTools MCP](https://ai-powered.com/mcp/chrome-devtools-mcp-3): Chrome DevTools MCP is an open-source MCP server for browser automation.
- [Selenium Webdriver MCP](https://ai-powered.com/mcp/selenium-webdriver-mcp): Selenium Webdriver MCP is an open-source MCP server for browser automation.

---
Source: https://ai-powered.com/mcp/mcp-server-browser
