browser-use/browser-use
Open-source browser automation agent for LLMs, usable as a Python library, CLI, hosted cloud service, or MCP server exposing browser-control tools.
Awesome MCP › Browser Automation
cloudflare/playwright-mcp
Cloudflare Playwright MCP is a port of the Playwright MCP server that runs as a Cloudflare Worker instead of a local process, using Cloudflare Browser Rendering and the @cloudflare/playwright package to drive the browser. Because the server is remote, clients connect to an SSE endpoint at the deployed worker URL rather than spawning a subprocess. Deployment is a short sequence: install dependencies with npm ci, build inside the cloudflare directory, then deploy the example worker with npx wrangler deploy; a Deploy to Cloudflare button is also provided. The README documents several client setups. In the Cloudflare AI Playground the worker URL is added under MCP Servers with the llama-3.3-70b-instruct-fp8-fast model, after which roughly fourteen tools are listed and the model can be asked to navigate, type and screenshot, with the advice to give one simple instruction at a time. Claude Desktop, which supports only local MCP servers, is connected through the mcp-remote proxy configured as an npx command in claude_desktop_config.json. VS Code and VS Code Insiders are configured with a single code --add-mcp command declaring an SSE server, making the tools available to the GitHub Copilot agent. For Cursor, inline image responses are disabled by default under the auto imageResponses setting, so a self-hosted worker must pass imageResponses set to allow when creating the MCP agent with createMcpAgent. Tools are offered in two modes: snapshot mode, the default, which uses accessibility snapshots for better performance and reliability, and vision mode, which uses screenshots and suits computer-use models that address elements by X and Y coordinates. Documented tools include browser_snapshot, browser_click with an optional double-click flag, browser_drag between two element references, browser_hover, browser_type and browser_navigate.
https://github.com/cloudflare/playwright-mcp
Also in Browser Automation
unclecode/crawl4ai
Open-source async web crawler and scraper that converts websites into clean, LLM-ready Markdown for RAG pipelines and AI agents, with an official MCP server for AI assistants.
microsoft/playwright-mcp
Playwright MCP is an MCP server that enables LLMs to automate and interact with web browsers using structured accessibility snapshots for efficient and deterministic browser automation.
alibaba/page-agent
Page Agent is an in-page JavaScript GUI agent that enables natural language control of web interfaces, offering easy integration without browser extensions or headless browsers.
Skyvern-AI/skyvern
Skyvern automates browser-based workflows using large language models and computer vision, providing a robust and scalable web automation solution that integrates with the Model Context Protocol (MCP).
browser-use/browser-harness
MCP-compatible harness that lets coding agents drive a user's real, logged-in Chrome browser over the DevTools Protocol, writing new helper scripts as it works.
apify/actors-mcp-server
Apify MCP Server is an MCP protocol server implementation that enables AI assistants to interact with Apify Actors for data extraction, web scraping, and automation tasks through a standardized protocol.
firecrawl/firecrawl-mcp-server
Firecrawl MCP Server is an MCP server implementation that adds advanced web scraping, crawling, and search capabilities to LLM clients like Cursor and Claude, enabling enhanced data extraction and research functionalities.