Hephaestus

Enables automation of Android devices and iOS simulators through a unified API. Supports device listing, screenshots with automatic compression for LLM processing, tap and swipe gestures by coordinates or element text, text input, app launching and installation, UI hierarchy inspection, device logs retrieval, and shell command execution. Provides cross-platform device control for mobile testing and debugging workflows.

Provides direct control over iOS simulators and physical iPhones through a unified interface that combines Facebook's idb for simulators and WebDriverAgent for physical devices. Features automated device discovery, real-time screen streaming via WebSocket for physical devices, and comprehensive interaction tools including tap, swipe, text input, button presses, screenshot capture, app launching, and UI element scanning. Supports both global and project-scoped setup with auto-configuration for Claude Code, Cursor, Codex, and OpenCode.

WeChat Mini-Program Agent enables AI-driven automation and testing of WeChat Mini Programs through a wrapper around miniprogram-automator's WebSocket interface to WeChat DevTools. Supported operations include page navigation, screenshot capture, UI element interaction, DOM inspection, and scenario testing. A serial execution model ensures predictable test flows for agent-first development workflows.

MCP server for Chrome browser automation via a companion Chrome extension. Provides 50+ tools covering navigation, element clicking, text input, screenshot capture, touch gesture simulation, macro recording and replay, DevTools inspection, and accessibility auditing. Communicates with the extension via a local WebSocket bridge at localhost:7890. Written in TypeScript, Apache 2.0 licensed.

Computer Control MCP provides computer automation capabilities through a server that enables AI assistants to control mouse movements, keyboard input, take screenshots, perform OCR, and manage windows. Built with PyAutoGUI, RapidOCR, and ONNXRuntime, it offers similar functionality to Anthropic's 'computer-use' tool but with zero external dependencies. The implementation supports operations like clicking, typing, dragging, capturing screen content with text recognition, and window management, making it particularly useful for automating desktop tasks and enabling AI systems to interact directly with graphical user interfaces.

BrowserControl is a browser automation server that enables interaction with web pages through a Set of Marks (SoM) system, where screenshots display numbered interactive elements for precise clicking and interaction. Built using Playwright and FastMCP, it provides persistent browser sessions with automatic cookie and login state management, developer tools integration for console logs and network monitoring, session recording capabilities, and smart navigation that handles localhost connection issues. The implementation is useful for web scraping, form filling, testing workflows, or any task requiring visual web interaction with debugging capabilities.

MCP Remote macOS Server provides secure remote control of macOS machines through VNC protocol, enabling AI assistants to capture screenshots and interact with the remote desktop through mouse and keyboard inputs. The implementation supports Apple Authentication (protocol 30) for secure connections, handles screen scaling for consistent display, and includes comprehensive error handling for connection issues. Built with a focus on security and reliability, this server is ideal for remote administration, automated testing, or AI-powered interaction with macOS systems.

Provides a unified interface for AI-assisted Godot Engine 4.x game development. Features include live editor manipulation of nodes, properties, and scenes, deep project context analysis of scripts and scene structures, end-to-end game testing with simulated inputs, automatic screenshot capture for visual context, bug fix loops with stack trace analysis, and version-aware documentation queries. Consolidates multiple AI-to-Godot features into a single toolkit with robust WebSocket connections and universal asset resolution.

Run a local Node.js server that receives WebP browser screenshots from the SnapStack browser extension, stores them at ~/.snapstack/, and exposes them to MCP clients. Provides tools to list, count, and clear pending captures, with all processing staying on localhost for privacy.

Improves frontend designs through screenshot-grounded iteration. Captures rendered pages, evaluates visual quality against design goals, and produces improved patches with proof artifacts including before/after screenshots and deterministic checks.

Webpage Screenshot MCP server enables AI assistants to capture visual screenshots of web pages using Puppeteer, with specialized support for authenticated workflows and multi-step interactions. Developed by Anand Tyagi, it provides tools for full-page and element-specific screenshots, manual login sessions with cookie persistence, and browser session management across multiple screenshot operations. The implementation supports both headless automation and visible browser modes, can integrate with system default browsers for natural user interactions, and handles authentication flows by allowing manual login followed by automated screenshot capture using saved cookies, making it ideal for monitoring protected web applications, visual verification of development progress, and automated testing workflows that require human authentication steps.

Provides computer use capabilities through an isolated Ubuntu desktop sandbox environment, implementing Anthropic's computer use API for safe AI agent interaction with desktop applications. Built on the BoxLite sandboxing library, it exposes a unified 'computer' tool that enables mouse control, keyboard input, screenshot capture, and scrolling within a 1024x768 XFCE desktop environment. Designed for AI agents that need to interact with graphical applications, browse the web, or manipulate files while maintaining complete isolation from the host system through hardware-level VM virtualization.

A fast screenshot capture tool optimized for AI vision models, particularly Claude Vision API, that automatically tiles full web pages into 1072x1072 pixel chunks for optimal processing. Built with Puppeteer and Sharp, it features intelligent browser lifecycle management with automatic restarts, memory optimization through page cleanup, and robust error handling with retry mechanisms. The server provides both single screenshot capture and screencast functionality with JavaScript execution capabilities, making it useful for web automation, visual testing, content analysis, and AI-assisted web development workflows where consistent, high-quality screenshots are needed.

MCP Server Pentest is a browser automation tool that enables AI assistants to perform security testing for web applications, specifically focusing on XSS and SQL injection vulnerability detection. Built on Playwright and Firefox, it provides tools for automatically testing URL parameters against common attack vectors, capturing screenshots of vulnerable pages, and manipulating web elements through various interaction methods. The server includes comprehensive error handling with fallback mechanisms for strict mode violations, making it particularly valuable for security professionals who need to conduct preliminary vulnerability assessments without leaving their AI assistant interface.

Mac Control provides 63 MCP tools for macOS automation written in native Swift with no external runtime dependencies. Tools cover screen capture, OCR, mouse and keyboard input, Windows UI Automation, Safari and Chrome browser automation via accessibility API, clipboard management, Spotlight search, file dialog control, and window/application management. Distributed as a signed, notarized macOS application bundle requiring no external runtimes.

Computer Use MCP enables AI agents to drive a local desktop through mouse clicks, keyboard input, scrolling, dragging, and screen capture. Written in Rust with cross-platform support for Windows, Linux (Wayland), and macOS 14+. Screenshots are downscaled by default, with automatic coordinate remapping back to actual desktop resolution. Exposes tools either as a single batched computer_use action or as individual split tools, matching the OpenAI Computer Use Agent specification.

Captures the frontmost macOS application, window title, browser URL every 5 seconds, and screenshots every 30 seconds, classifying activities using a locally-running Ollama model. Builds a searchable record of people, documents, links, and tickets encountered during the workday. An MCP server lets Claude query this local activity history with natural language. All processing is fully local with no cloud dependencies or telemetry.

This PageSpeed MCP server, developed by Phiality, integrates Google's PageSpeed Insights API with AI assistants to enable comprehensive web performance analysis. Built with TypeScript and leveraging the Model Context Protocol SDK, it provides detailed metrics on Core Web Vitals, best practices, SEO, accessibility, and resource optimization. The server implements error handling, custom configurations, and uses a standardized JSON response structure. By abstracting the complexities of performance auditing, it allows AI systems to easily analyze and interpret website performance data. This implementation is particularly valuable for applications requiring in-depth web performance insights, facilitating use cases such as SEO optimization, user experience improvement, and technical website audits.

The macOS GUI Tools server provides AI assistants with the ability to interact with native macOS applications through a set of specialized tools. Built with Python using the FastMCP framework, it enables operations like capturing screenshots of applications, identifying UI elements (buttons and text areas), moving the mouse cursor to specific coordinates, clicking, and typing text. The implementation supports various Apple applications including Notes, Safari, Mail, and others by leveraging the macOS accessibility framework to extract UI element information. This server is particularly valuable for workflows requiring desktop automation within AI assistant conversations, allowing users to control their Mac applications without manual intervention.

Runs locally and gives AI models web-browsing capabilities including internet search, full-page screenshots, and clean text extraction from websites. Designed for local LLMs running in environments like SillyTavern with Gemma or Qwen models. Accessible as an HTTP endpoint at localhost:4242/mcp.

This MCP server implementation provides a bridge to Cloudflare's Browser Rendering API, enabling web scraping and screenshot capture capabilities. It utilizes Puppeteer for browser automation and offers endpoints for content extraction and screenshot generation. The server is designed for use cases requiring dynamic web content processing, such as data extraction from JavaScript-heavy websites or automated visual testing of web applications.

This MCP server provides browser automation capabilities through Chrome's debugging protocol with session persistence, enabling AI agents to control Chrome instances for web scraping, testing, and automation tasks. Built using TypeScript with puppeteer-core and the Model Context Protocol SDK, it offers tools for taking screenshots in multiple formats (PNG, JPEG, WebP), navigating to URLs, clicking elements, typing text, and retrieving page content while maintaining browser sessions across multiple interactions. The implementation features automatic browser discovery, connection management with configurable debugging ports, error handling with detailed responses, and support for both headless and headed Chrome instances, making it valuable for automated testing workflows, web data extraction, and building AI assistants that need persistent browser control without manual Chrome DevTools access.

Open-source Windows MCP server for AI assistants. File, clipboard, screenshot, process & power mgmt.

DomShot MCP server enables AI assistants to capture screenshots of specific DOM elements through an interactive browser interface. The implementation combines a Go-based MCP server with a JavaScript client that injects an overlay system allowing users to hover over and click on webpage elements to generate PNG screenshots using html2canvas. It operates by serving a test HTML page and WebSocket connection on localhost:8000, where the browser-side script communicates screenshot requests back to the MCP server, making it useful for visual documentation, UI testing, or creating element-specific captures for design workflows.

BrowserCat MCP Server provides cloud-based browser automation capabilities through the Model Context Protocol, enabling LLMs to interact with web pages without local browser installation. Developed by BrowserCat, it offers tools for navigation, screenshot capture, element interaction (clicking, hovering, form filling), and JavaScript execution in a real browser environment. The server connects to BrowserCat's cloud browser service via WebSocket, authenticates with an API key, and exposes both console logs and screenshots as resources. This implementation is particularly valuable for AI assistants that need to perform web-based tasks like data extraction, form submission, or visual analysis of websites.

Gives AI agents full desktop control on macOS through 18 tools covering screen capture, mouse and keyboard input, window and application management, UI element inspection via the Accessibility API, and clipboard operations. Uses a pre-compiled Swift binary for system interactions and AppleScript for keyboard input, with no native Node.js addons or build steps required. Supports macOS 13 and later on both Intel and Apple Silicon architectures.

Bridges Flutter mobile app development with Maestro UI testing by providing tools for inspecting UI hierarchies, simulating user interactions (tap, text input, scrolling), taking screenshots, and managing app state through hot reload/restart. Connects to running Flutter apps via VM Service to enable real-time testing and automation workflows, while also supporting native Maestro YAML flow creation and execution. Includes a complete example Flutter counter app with TestHooks for isolated testing.

This MCP server provides comprehensive Lighthouse performance analysis capabilities through automated web performance auditing and deep analysis tools. Built with TypeScript and featuring integration with Google's Lighthouse engine, Puppeteer, and Chrome launcher, it offers four core tools: URL analysis with configurable device types and audit categories, deep analysis with critical request chain examination and unused code detection, batch analysis for multiple URLs in parallel, and prompt generation for AI-assisted performance optimization workflows. The implementation includes specialized analyzers for Core Web Vitals, performance patterns, problem detection, and score breakdowns, with browser pool management for efficient parallel processing and comprehensive reporting that generates prioritized recommendations with impact assessments, making it ideal for web developers, performance engineers, and automated CI/CD pipelines that need programmatic access to detailed website performance insights and optimization guidance.

Exposes Abrasio's browser automation capabilities to AI agents with built-in anti-detection fingerprinting. Provides 15 tools covering navigation, screenshots, clicking, form input, and JavaScript execution. Supports local operation using a standard browser and cloud operation with residential IP infrastructure. Installable via pip.

Extracts text from PDFs in LLM-friendly formats by preserving table structure and detecting figure-heavy pages using spatial heuristics tuned on real-world documents. Renders page screenshots locally without requiring GPU, OCR, or external API keys. Built in TypeScript with a local-only deployment model.

Kimi WebBridge enables AI agents to control a browser through the Kimi WebBridge Chrome extension, exposing tools for navigation, element clicking, form filling, screenshot capture, and JavaScript execution. Requires the companion Chrome extension to be installed. Installable via npx, stdio transport. Written in TypeScript.

Provides visual feedback for frontend development by giving coding agents direct access to rendered output. Uses Puppeteer to capture screenshots, collect console logs, track network errors, and perform visual regression detection via pixel comparison. Supports browser interactions including click, type, scroll, hover, form selection, and viewport resizing with automatic screenshot capture after each action.

URL2SNAP is an MCP server that captures screenshots of websites through the Abstract API. It provides a simple tool that takes a URL as input and returns a local URL to access the captured screenshot. The implementation handles API requests, image processing, and local file storage, making it useful for applications that need to visualize web content without requiring users to visit the actual websites. The server runs on port 8011 and saves screenshots to a local directory for easy access.

Connects to a local PinchTab server for headless browser control, providing 16 tools for navigating pages, clicking elements, filling forms, taking screenshots, and more. Uses PinchTab's accessibility-tree snapshots for token-efficient page understanding, with all interactions using stable element references instead of CSS selectors. Automatically starts and manages the local PinchTab server process, requiring only PinchTab to be installed globally.

Provides browser automation capabilities through Puppeteer, enabling web page interaction, content extraction, screenshot capture, PDF generation, and JavaScript evaluation for web scraping, automated testing, and form filling workflows.

Illustrator MCP Server enables AI assistants to execute JavaScript/ExtendScript code directly in Adobe Illustrator and view the results. Developed by Anshul Dalua, this MacOS-only implementation uses AppleScript to bridge between the MCP protocol and Illustrator's scripting capabilities, allowing bots to programmatically generate and manipulate vector graphics. The server provides two main tools: one for running arbitrary code in the current Illustrator document and another for capturing screenshots of the Illustrator window to visualize results, making it particularly useful for automated design tasks and graphic generation workflows.

A browser automation MCP server built with Playwright that provides web interaction capabilities through standardized tools for navigation, content extraction, form filling, screenshot capture, and JavaScript execution. The implementation uses a multi-stage Docker setup with proper resource management and security configurations, supporting both headless and headed browser modes with configurable timeouts and viewport settings. Designed for AI assistants that need reliable web automation capabilities, it handles common browser tasks like navigating to URLs, extracting page content with optional CSS selectors, clicking elements, filling forms with automatic submission, and capturing screenshots of full pages or specific elements.

CatchMetrics provides real user monitoring (RUM) for web performance optimization. The platform tracks Core Web Vitals (LCP, CLS, INP), navigation timings, JavaScript long tasks, and third-party script impact across devices and regions. AI-powered analysis helps identify performance degradation causes and correlates metrics with deployment history to enable faster issue resolution.

GTmetrix exposes its web performance analysis platform through an MCP server, enabling AI agents to run speed tests, retrieve PageSpeed and YSlow scores, access Core Web Vitals, and monitor website load times. Results include waterfall charts, performance recommendations, and historical comparisons. Requires a GTmetrix API key.

SpeedVault's official MCP server for web performance and SEO. Runs Lighthouse-based performance audits, provides code-level fix recommendations, crawls pages for SEO analysis, and reads real-user Core Web Vitals data. Requires a SpeedVault API key.

Petamind is an agentic coding loop MCP server that generates, tests, and visually evaluates code modifications. It provides multi-candidate patch generation with deterministic gates, integrating quality checks including accessibility testing via axe-core and performance benchmarks via Lighthouse. The system uses Playwright for screenshot capture and supports various LLM providers (Vertex AI, OpenRouter) for patch generation and vision-based assessment of UI changes.

Captures screenshots of any web page in PNG, JPEG, or WebP format and generates PDFs with configurable options. Supports full-page capture, custom viewport sizes, and output format selection. Uses x402 USDC micropayments on Base for pay-per-call access.

Browser Proof provides AI agents with verifiable web content evidence through deterministic JSON capture, text extraction, and screenshot generation. It enables agents to produce tamper-evident proof of web page states, supporting fact-checking and evidence collection workflows.