How Midscene MCP Empowers AI Assistants to Automate Browser Tasks

Midscene MCP Overview

Midscene provides MCP services that allow AI assistants to control browsers via natural‑language commands, automate UI tasks, and generate Midscene automation scripts.

Use Cases

Control browser to perform automation tasks

Generate Midscene automation scripts

Usage Example

Generate a Midscene test case for the Sauce Demo site.

Setup Midscene MCP

Prerequisites

OpenAI API key or another supported AI model provider.

For Chrome integration (bridge mode): install the Midscene Chrome extension and switch the extension to "Bridge mode" then click "Allow connection".

Configuration

Add the Midscene MCP server to your MCP configuration:

{
  "mcpServers": {
    "mcp-midscene": {
      "command": "npx",
      "args": ["-y", "@midscene/mcp"],
      "env": {
        "MIDSCENE_MODEL_NAME": "REPLACE_WITH_YOUR_MODEL_NAME",
        "OPENAI_API_KEY": "REPLACE_WITH_YOUR_OPENAI_API_KEY",
        "MCP_SERVER_REQUEST_TIMEOUT": "800000"
      }
    }
  }
}

For more information on configuring the AI model, see the "Choose AI Model" reference.

Available Tools

Navigation : midscene_navigate – navigate the current tab to a specified URL.

Tab Management : midscene_get_tabs – list all open tabs; midscene_set_active_tab – switch to a tab by ID.

Page Interaction : midscene_aiTap – click an element described in natural language; midscene_aiInput – input text into a field; midscene_aiHover – hover over an element; midscene_aiKeyboardPress – press a keyboard key; midscene_aiScroll – scroll the page or a specific element.

Verification & Observation : midscene_aiWaitFor – wait for a condition to become true; midscene_aiAssert – assert a condition; midscene_screenshot – capture a screenshot of the current page.

Playwright Example : midscene_playwright_example – provides a Playwright code example.

FAQ

What advantages does Midscene MCP have over other browser MCPs?

Supports Bridge mode, allowing direct control of the current browser without re‑login or downloading a new browser.

Includes optimal prompt templates and execution practices for a more stable and reliable automation experience.

Automatically generates execution reports that can be viewed after each task.

Local port conflict when multiple clients run

Problem description

When several clients (Claude Desktop, Cursor MCP, etc.) use Midscene MCP simultaneously, the server port may be occupied, causing errors.

Solution

Close the extra MCP server instances.

Run the following commands to free the port:

# For macOS/Linux:
lsof -i:3766 | awk 'NR>1 {print $2}' | xargs -r kill -9

# For Windows:
FOR /F "tokens=5" %i IN ('netstat -ano ^| findstr :3766') DO taskkill /F /PID %i

How to obtain the Midscene execution report

After each task, an HTML report is generated and can be opened directly from the command line.

# Replace with your report file name
open report_file_name.html

References

MCP: https://modelcontextprotocol.io/introduction

Choose AI Model: https://midscenejs.com/zh/choose-a-model

Chrome Web Extension: https://chromewebstore.google.com/detail/midscenejs/gbldofcpkknbggpkmbdaefngejllnief?hl=zh-CN&utm_source=ext_sidebar