sjawhar/browser-mcp

Browser MCP

Automate browser interactions via BrowserMCP. Use skill_mcp(mcp_name="browsermcp", ...) to invoke tools.

Requires the BrowserMCP Chrome extension to be installed and connected.

Tools

| Tool | Purpose | | ------------------------ | ------------------------------------------------ | | browser_navigate | Navigate to a URL | | browser_go_back | Go back to the previous page | | browser_go_forward | Go forward to the next page | | browser_snapshot | Capture accessibility snapshot of current page | | browser_click | Click an element on the page | | browser_hover | Hover over an element | | browser_type | Type text into an editable element | | browser_select_option | Select option(s) in a dropdown | | browser_drag | Drag and drop between two elements | | browser_wait | Wait for a specified number of seconds | | browser_press_key | Press a keyboard key | | browser_screenshot | Take a PNG screenshot of the current page | | browser_get_console_logs | Get browser console log output |

Interaction Pattern

Most interaction tools (browser_click, browser_type, etc.) require an element + ref pair:

• ref — opaque element ID from a browser_snapshot accessibility tree
• element — human-readable description of the element (used for permission display)

Always call browser_snapshot first to get valid ref values. Refs are page-session-specific and change on navigation.

Navigation and interaction tools automatically return a fresh snapshot after executing — no need to call browser_snapshot separately after them.

Usage Examples

skill_mcp(mcp_name="browsermcp", tool_name="browser_navigate", arguments='{"url": "https://example.com"}')

skill_mcp(mcp_name="browsermcp", tool_name="browser_snapshot")

skill_mcp(mcp_name="browsermcp", tool_name="browser_click", arguments='{"element": "Login button", "ref": "e42"}')

skill_mcp(mcp_name="browsermcp", tool_name="browser_type", arguments='{"element": "Search input", "ref": "e15", "text": "hello world", "submit": true}')

skill_mcp(mcp_name="browsermcp", tool_name="browser_select_option", arguments='{"element": "Country dropdown", "ref": "e88", "values": ["US"]}')

skill_mcp(mcp_name="browsermcp", tool_name="browser_screenshot")

skill_mcp(mcp_name="browsermcp", tool_name="browser_press_key", arguments='{"key": "Escape"}')

skill_mcp(mcp_name="browsermcp", tool_name="browser_wait", arguments='{"time": 2}')

Typical Workflow

1. browser_navigate to a URL (returns snapshot)
2. Read the snapshot to find element ref values
3. Interact with elements using browser_click, browser_type, etc.
4. Each interaction returns an updated snapshot — use it for the next step
5. Use browser_screenshot when you need a visual capture

Notes

• The BrowserMCP Chrome extension must be active and connected — without it, all tools error with "No connection to browser extension"
• browser_screenshot returns a PNG image (base64), not text
• browser_get_console_logs returns JSON-stringified log entries, one per line
• browser_drag requires both start and end element/ref pairs (startElement, startRef, endElement, endRef)
• Key names for browser_press_key follow web standards: Enter, Tab, Escape, ArrowLeft, ArrowDown, a, A, etc.