Browserbeam MCP Server
Provides browser automation tools via Browserbeam, enabling AI agents to navigate, click, fill forms, extract data, and perform other browser actions directly in Windsurf (Codeium IDE).
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Browserbeam MCP ServerNavigate to hackernews and list the top stories"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Browserbeam MCP Server
MCP (Model Context Protocol) server for Browserbeam — use real browser automation as tools in Cursor, Claude Desktop, Windsurf, and any MCP-compatible client.
Setup
Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"browserbeam": {
"command": "npx",
"args": ["-y", "@browserbeam/mcp-server"],
"env": {
"BROWSERBEAM_API_KEY": "sk_live_your_key_here"
}
}
}
}Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"browserbeam": {
"command": "npx",
"args": ["-y", "@browserbeam/mcp-server"],
"env": {
"BROWSERBEAM_API_KEY": "sk_live_your_key_here"
}
}
}
}Windsurf
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"browserbeam": {
"command": "npx",
"args": ["-y", "@browserbeam/mcp-server"],
"env": {
"BROWSERBEAM_API_KEY": "sk_live_your_key_here"
}
}
}
}Related MCP server: Cloudflare Playwright MCP
Available Tools
Tool | Description |
| Create a browser session, optionally navigate to a URL |
| Navigate to a new URL in an existing session |
| Get page content as markdown or HTML with interactive element refs. Supports |
| Click an element by ref, text, or label |
| Fill form fields or an entire form at once |
| Type text character-by-character with real keyboard events |
| Select an option from a dropdown |
| Check or uncheck a checkbox or radio button |
| Scroll the page or scroll an element into view |
| Scroll the entire page to load lazy content, then observe |
| Wait for a selector, text, JS expression, or fixed delay |
| Extract structured data using a declarative schema |
| Run custom JavaScript in the browser page context |
| Take a screenshot of the current page |
| Generate a PDF of the current page |
| Upload files to a file input element |
| List your sessions; optional |
| Get the status, duration, and error fields (e.g. |
| Close a session and release resources |
Page Map & Full Mode
The first observe in every session auto-includes a page map — a lightweight outline of page sections (nav, header, main, aside, footer) with CSS selectors and content hints. This lets agents discover what's on the page beyond the main content area without spending tokens.
To get content from all page sections instead of just the main area, use mode: "full":
{
"tool": "browserbeam_observe",
"params": {
"session_id": "ses_abc123",
"mode": "full",
"max_text_length": 20000
}
}The response organizes content by section:
## [nav]
Home | Products | About | Contact
## [main]
# Welcome to Our Site
...main content...
## [aside]
Related links, sidebar widgets...
## [footer]
© 2026 Company | Privacy | TermsUse include_page_map: true to re-request the page map on subsequent observations.
Proxies
All sessions use a datacenter proxy by default (country auto-detected from the URL's TLD). To customize, pass proxy_kind and/or proxy_country to browserbeam_create_session:
{
"tool": "browserbeam_create_session",
"params": {
"url": "https://example.com",
"proxy_kind": "residential",
"proxy_country": "us"
}
}AI-Powered Selectors
Use the ai >> prefix in extract schemas to describe fields in plain English. The engine resolves them to CSS selectors via AI and caches the result:
{
"tool": "browserbeam_extract",
"params": {
"session_id": "ses_abc123",
"schema": {
"_parent": "article.product_pod",
"name": "ai >> the product title",
"price": "ai >> the price including currency symbol"
}
}
}Agent guidelines (for AI clients)
Close sessions: Agents should call
browserbeam_closewhen finished with a session so resources are released and credit consumption stops. Only keep a session open if the user explicitly needs continued work on the same browser.Page discovery: The first observe auto-includes a
map. Check it before usingmode: "full"— if the info you need is in the main content, default mode is more token-efficient.Full mode: Use
mode: "full"when you need sidebar content, footer links, or navigation items that aren't in the main area. Defaultmax_text_lengthfor full mode is 20,000 characters.Truncation: Page markdown is capped by default at 12,000 characters (
browserbeam_observeand the page payload frombrowserbeam_create_session/browserbeam_navigate). If output is truncated, usebrowserbeam_observewith a highermax_text_lengthorbrowserbeam_scroll_collect(default 100,000 characters) for long or lazy-loaded pages.
How It Works
The MCP server translates tool calls into Browserbeam API requests. Your AI agent sees structured page data (markdown content, interactive element refs, change diffs) instead of raw HTML.
AI Agent → MCP Tool Call → Browserbeam API → Real Browser → Structured Response → AI AgentEnvironment Variables
Variable | Required | Description |
| Yes | Your Browserbeam API key ( |
| No | API base URL (default: |
Get an API Key
Sign up at browserbeam.com — 5,000 free credits, no credit card required.
License
MIT
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/nyku/browserbeam-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server