MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
A browser body for an AI agent: drive a real, persistent Chrome over MCP. Accessibility-tree perception, act-by-name, no CDP.
MIT ★ 1 -
browse-mcpgithub.com/That1Drifter/browse-mcpupdated this year
Headless-browser MCP server for any MCP-capable agent (Claude, Codex, Gemini CLI, Cursor, ...). Playwright-based: accessibility-tree refs, Readability + PDF extraction, search, research macro, CLI mode. Windows / Linux / macOS.
npm MIT ★ 1 -
browse-mcpgithub.com/that1drifter/browse-mcpupdated this year
Headless-browser MCP server for any MCP-capable agent (Claude, Codex, Gemini CLI, Cursor, ...). Playwright-based, with accessibility-tree refs, Readability, search, annotated screenshots, and a self-improvement feedback loop. Windows, Linux, and macOS.
npm MIT -
Diwall MCP — the operator's browser, seen and handled by a language model. An MCP server that gives the model the screenshot, the accessibility tree and the gestures of the browser where the operator is signed in, under their confirmation.
MIT -
Computer use for Grok Build: a plugin and MCP server that let the agent see the screen, click and type in desktop apps and browsers on macOS, Windows and Linux. Accessibility-tree first, with safety hooks. Built for UI testing in the dev loop.
Apache-2.0 -
browse-nowgithub.com/nowledge-co/nowledge-memupdated this year
Browser automation CLI for AI agents - reliable element targeting through accessibility tree refs
pypi Proprietary -
High-performance Rust MCP server for Windows desktop automation, native accessibility tree navigation, and system-level LLM orchestration.
MIT -
pinchtab-mcpgithub.com/maderwin/pinchtab-mcpupdated this year
MCP server for PinchTab browser automation — control Chrome via accessibility tree snapshots
npm MIT -
mcp-linux-desktopgithub.com/wizardofweb125-lab/mcp-linux-desktopupdated this year
MCP server for full Linux desktop control — windows, mouse, keyboard, clipboard, audio, screenshots, OCR, notifications, accessibility tree.
MIT ★ 1 -
screenreadgithub.com/Bambushu/screenreadupdated this year
Your AI agent doesn't need screenshots to read text. ScreenRead gives it the macOS accessibility tree in ~100ms. CLI + MCP server.
MIT ★ 3 -
Claude Code computer use for Windows and macOS through the accessibility tree, not screenshots.
MIT ★ 1 -
Claude-style computer use for any vision model in Cursor on macOS (MCP server: screenshots + accessibility tree + OCR)
★ 1 -
Computer use for Claude Code on Windows and macOS through the accessibility tree.
npm MIT -
macpilot-mcpgithub.com/passerby2049/macpilot-mcpupdated this year
MCP server that lets Claude Code drive macOS apps — screenshots, accessibility tree, clicks, keyboard, and menu bar. Inspired by Lakr233/ComputerUse.
-
Local desktop automation for MCP-capable agents on macOS, Linux and Windows. Screenshots, accessibility trees and input stay on your machine.
MIT -
Accessibility-first computer-use framework for macOS, Windows, Linux, and the browser (CDP): pruned accessibility-tree snapshots, element refs, and a safety layer, exposed as an MCP server + CLI.
pypi Apache Software License -
Fast macOS computer-use CLI & MCP server for AI agents (Claude, Codex, ChatGPT, Grok, DeepSeek): screenshots, mouse, keyboard, accessibility tree, GUI automation. Zero-dependency Swift.
MIT ★ 1 -
MCP server letting AI agents drive a Linux/X11 desktop — AT-SPI accessibility-tree perception with element-level actions, instead of screenshot + coordinate clicking.
MIT -
Let AI agents use your Mac's apps in the background: accessibility-tree computer use as an MCP server. Native Swift, no dependencies.
MIT -
Let AI agents use your real Android phone through its accessibility service: screen tree with stable refs, taps, typing, gestures, notifications, apps. MCP server + hub + app.
MIT ★ 5 -
computer use for ai agents on omarchy and hyprland: screenshots, mouse, keyboard, ocr, accessibility tree and window control. a claude code plugin and mcp server.
MIT -
pc-screen-controlgithub.com/desteny-dev/pc-screen-controlupdated this year
MCP server for Windows. Gives an AI the accessibility tree instead of screenshots, so it operates controls by name and reports the state before and after. 34 tools, no network access, one-click install for Claude Desktop, works with any MCP client.
MIT ★ 2 -
a11y-computer-use: accessibility-first computer use for AI agents. Pruned a11y-tree snapshots, element refs instead of pixel coordinates, a safety layer, an MCP server and CLI. macOS, Windows, Linux, and Chromium over CDP.
Apache-2.0 ★ 1 -
Lets AI agents see and control a desktop GUI: screenshots, OCR, mouse, keyboard, windows and accessibility-tree UI elements. One-shot JSON CLI or MCP server. Windows 10/11, and Linux on KDE Plasma 6 or GNOME 45+ (Wayland).
MIT ★ 1 -
DeepSeek Harness plugin: migrates OpenAI Codex's chrome@openai-bundled plugin into DSH as native mcp__chrome__* tools — drive your own Chrome (tabs, logged-in sessions, accessibility trees, screenshots, input) with a bundled control-chrome skill and zero runtime dependencies.
MIT -
Computer use MCP server for Windows: accessibility tree + pixels, code mode, game-grade input and a safe takeover overlay
MIT ★ 1 -
background-computer-usegithub.com/Panchangam18/background-computer-useupdated this year
MCP server + skill library for driving macOS apps in the background — click, type, scroll, and inspect accessibility trees without activating apps or moving your cursor. Works with Claude Code, Claude Desktop, Cursor, Codex CLI, Goose, and any other MCP-capable LLM client.
MIT ★ 5 -
pinchtab-mcpgithub.com/maderwin/pinchtab-mcpupdated this year
MCP server for PinchTab — control Chrome (or any Chromium browser) via accessibility tree snapshots through the Model Context Protocol.
MIT -
Windows MCP server: AI sees and drives any app through the accessibility tree - no screenshots, no mouse - plus 200+ native Windows tools.
Apache-2.0 ★ 3 -
Reliable, efficient Mac control for AI agents: an MCP server that drives macOS apps through the accessibility tree, with verified actions, a safety policy and an audit log.
MIT -
Eyes, ears and hands for AI agents on GUI programs: a sealed X11 or Wayland session where an agent (over MCP), a script or CI reads the accessibility tree, takes screenshots, records video and sound, clicks, types and drives gamepads, then checks contrast, Tab order, visual diffs, flicker, input latency and audio dropouts. CLI, Python, Playwright.
npm MIT ★ 1 -
AX-first computer use for AI agents on macOS with optional Jev (TypeSafe System One) semantic guards: calibrated target/input judgments before an irreversible action, decisions kept in code. Accessibility-tree targeting, window-scoped input, clipboard-safe paste, read-back verification. Ships a pip CLI, a pi package and a DeepSeek Harness plugin.
MIT ★ 8 -
A MCP server that bridges AI agents to the macOS desktop: the accessibility tree, synthetic mouse and keyboard input, screenshots, window and application control, the shell, preferences, launchd, processes, the filesystem and web scraping.
MIT -
selenium-ai-agentgithub.com/learn-automated-testing/selenium_agentupdated this year
AI-powered Selenium MCP server for browser automation — 74 tools with accessibility tree discovery, test generation, self-healing, and Selenium Grid parallel execution for AI agents
npm MIT -
An MCP server that lets Claude Code use your Windows apps. It reads each window as text through the accessibility tree, presses, fills and selects controls through UI Automation without moving your mouse, and replies to every action with what changed. One exe under 200 KB, no admin rights, free (Apache-2.0). Tested on Windows 11 with Claude Code.
Apache-2.0 -
cdpilotno update data
cdpilot is a browser automation CLI and local MCP server that talks to Chrome, Chromium, Brave or Vivaldi directly over the Chrome DevTools Protocol, with no Puppeteer, Playwright or Selenium and no browser download. Its 45 MCP tools cover navigation, clicks and form filling (including inside cross-origin iframes), accessibility-tree snapshots, screenshots, test assertions, a page's own WebMCP tools, and Web Bot Auth request signing. `cdpilot connect` attaches to a browser you started yourself, so a person can clear a login or CAPTCHA and the agent continues; cdpilot never closes that browser. Every feature has a runnable example with its real output: https://github.com/mehmetnadir/cdpilot/tree/main/examples
smithery -
An MCP server that drives a real Android device over ADB and attaches the resulting screen to every action. Single native binary, continuous H.264 vision cache, OCR for screens the accessibility tree cannot read.
MIT ★ 4 -
MCP server for evidence-disciplined genealogy: read/write access to a self-hosted Gramps Web tree where every fact carries a citation, plus a read-only reference layer over legacy GEDCOM exports.
MIT -
Alcove is an MCP server that gives AI coding agents on-demand access to your private project docs — BM25 + vector hybrid search for precision retrieval, tree-sitter code indexing so agents understand your codebase structure, and policy enforcement for doc consistency.
Apache-2.0 ★ 16 -
A MCP server that bridges AI agents to the Windows desktop. UI Automation, synthetic mouse/keyboard input, screenshots, window and application control, PowerShell, the registry, processes, the filesystem, and web scraping. No computer-vision model is required: the agent perceives the UI through the Windows accessibility tree.
MIT ★ 17 -
Read-only MCP server for Trino: each query runs as the user who asked, and Trino's access control decides what they see. Nine tools, writes refused on the SQL tree, bounded SQL, rows, bytes and time. Any MCP host, any model.
pypi Apache-2.0 ★ 7 -
Browser automation MCP server with event-driven deterministic snapshots (Playwright + a11y tree) / 事件驱动确定性快照的浏览器操控 MCP 服务器
MIT -
Cross-platform GUI MCP for interacting with your desktop in macos, Windows, and Linux!
pypi MIT ★ 8 -
site-sensegithub.com/NordmanLive/site-senseupdated this year
Give your AI coding CLI eyes into web portals. No cloud. No storage. Read-only.
Apache-2.0 ★ 2 -
Hands your live Chrome tab to any MCP client — snapshot, navigate, click, type, extract.
npm MIT -
Official code for "Hybrid Computer Use: A Unified Programmatic, Semantic, and Visual Interface for General Agents"
MIT ★ 55 -
MCP server that lets Claude Code, opencode and other agents use your Mac apps in the background, without moving your mouse or taking focus. Built on agent-desktop.
Apache-2.0 ★ 2 -
browser-cognition-mcpgithub.com/hrshx3o5o6/brocogniupdated this year
Brocogni -- Making Playwright make sense to AI agents. MCP server for semantic browser observation and self-healing selectors.
npm MIT -
AX-first macOS computer use for DeepSeek Harness: accessibility snapshots with stable refs, background input, Unicode typing, menus, apps, set-of-mark screenshots, on-device OCR, verified actions, and optional Jev (TypeSafe System One) semantic guards — 1
npm MIT -
Public Browser: New Chrome AI Tool Just Got 40% Faster Browser Automation! (Playwright Killer) - High-speed Chrome AI browser automation MCP server and Python Script API.
★ 2 -
webpilot-cligithub.com/luckysolanki902/webpilotupdated this year
A semantic terminal browser for LLM agents and CLI-native developers. Browse any website from your terminal — no GUI, no mouse, just commands.
npm MIT -
Zero-token ($0) E2E Playwright QA engine for Claude Code, Google Antigravity & Cursor. Crawls live DOMs into ~1.2k-token accessibility maps, runs self-healing intent plans locally, and compiles to native .spec.ts. Dual-mode: Agent Skill CLI + Stdio MCP Server.
MIT ★ 2 -
code-graph-mcpupdated this year
A code indexing system accessible via MCP (Model Context Protocol)
npm MIT -
code-graphgithub.com/ysterbal/code-graphupdated this year
Multi-language code parser (Python, Rust, JS/TS, Go, Java, C/C++) that extracts dependency graphs and stores them locally in SQLite. Provides three access methods: CLI for direct queries, REST API for programmatic use, and MCP server for IDE integration with Claude or Cursor.
★ 2 -
MCP server that lets Claude see and use the Windows desktop: UI Automation tree, annotated screenshots and real mouse/keyboard input, with input lock and emergency stop. EN/TR.
MIT ★ 7 -
directionmdgithub.com/Dragoon0x/directionmdupdated this year
Infer creative direction from any live URL. 16 extraction passes, 22 inference modules, decision trees, design tokens, competitive positioning, drift monitoring. One crawl → DIRECTION.md
NOASSERTION ★ 1 -
MCP server to drive Android from the UI tree only (no screenshots, no vision) — ~100-250 tokens per screen. Built on google/artemis.
Apache-2.0 -
Windows computer-use MCP server for OpenCode, Claude Desktop and Codex CLI. Indexed UI-Automation controls, set-based tree filtering, verified clicks - no vision model, no pixel guessing.
Apache-2.0 -
MCP server that lets an AI coding agent run and drive a real GPUI desktop app: screenshots, element tree, clicks, typing, scrolling, dialogs and file pickers, headless or in a sandboxed KDE Plasma session on Linux
Apache-2.0 ★ 2 -
Computer use MCP server for Claude Code, Codex and other AI agents. Reads the UI tree instead of taking a screenshot every step, with Jev (TypeSafe System One) as a fast judge that can also look at the screen to find icons and game objects. Local macOS, local Windows, or remote Windows over SSH.
MIT ★ 4