MCP servers · shortlist
MCP servers for browser and OS automation
Browser drivers, desktop control, smart-home bridges and embedded devices — servers that let an agent act on a machine instead of only reading from it.
How this list was ordered. There is no reliable popularity signal in the source registries, so entries are ranked by how completely the record is documented — a substantial description plus a documentation, repository or base URL — and then alphabetically. 113 mcp servers in this scope clear the content bar; the top 24 are shown. A place here is not an endorsement or a paid slot.
Shortlisted entries
- 01
ostapondo/Plonk
A Mac window manager an agent can drive, plus ten more menu bar utilities behind the same icon: see every monitor and window, arrange a layout across displays in one call, snap a window into zones you drew yourself, save workspaces that relaunch apps into their places, keep-awake, measure the screen, screenshot a window by name even when it is buried behind other windows, annotate it, and read text off the screen on-device — handed back a box per line, so what was read can be pointed at. 20 tools, backed by a native menu bar app over a local-only loopback API; no cloud, no telemetry. npx -y plonk-mcp, or open the shipped .mcpb bundle in Claude Desktop.
OS Automation auth: None
- 02
fixed-width/glass
Gives a coding agent a build → see → interact → debug loop over the native GUI app it is writing: launch the app, screenshot it, read its accessibility tree, inject mouse/keyboard/gestures, tail its logs, and diff frames to confirm a change landed. Drives apps as an external black box with no app integration, so any toolkit or language works. 30 tools with text-first responses (element ids, diffs, waits) so the agent need not screenshot every step, plus optional sandboxing of the launched app. Backends: X11, Wayland, Windows, macOS, Android emulators, and the iOS Simulator.
OS Automation auth: None
- 03
dmytrome/groundhog
Web search, read and research over a real stealth-patched Chrome, driven by raw CDP so the Runtime domain is never enabled and the isAutomatedWithCDP tell is absent. Strips text that is invisible in the rendered DOM before the model reads it and reports each occurrence in threats, returns a SHA-256 provenance receipt per source, and re-checks the SSRF guard after redirects. research ranks passages across all sources in one pass — extracts, not summaries, so no model or API key is involved. MIT.
Browser Automation auth: API Key
- 04
ilien-dev/svipall
Local-first web reading with a tier ladder that escalates from plain HTTP to a real browser when a page pushes back. Pages as clean Markdown, whole-site crawls, keyless search, repeated page structure induced into rows, and a REST API. Attempts supported captchas on your own machine with optional local ONNX models, and parks the rest in a human dashboard. Labels blocked and low-quality pages instead of handing back a challenge screen as content. No API keys, no paid solvers, no telemetry.
Browser Automation auth: API Key
- 05
SolveGate/solvegate-mcp
Deal with Cloudflare Turnstile from an agent. inspectpage reports whether a page actually uses Turnstile and what its sitekey is — free, no API key — so the agent can look before it acts, and it says plainly when a page carries reCAPTCHA or hCaptcha instead, or is a full-page Cloudflare interstitial rather than an embedded widget. solveturnstile and getsolve clear the challenge and return a token when one is genuinely needed. npx -y solvegate-mcp
Browser Automation auth: API Key
- 06
purinzan/gx3-cli-mcp
Read-only analysis of Mitsubishi Electric MELSEC PLC projects saved by GX Works3 (.gx3): trace why a coil never turns on through the ladder logic, find where a device is written and read, search by device comment, and separate conditions that arrive from physical inputs, HMI or network communication. Runs entirely locally and never modifies the project; no project-mutating command is exposed. Unofficial, not affiliated with Mitsubishi Electric.
Industrial & IoT auth: None
- 07
yinnho/aginxbrowser
A browser built for AI agents as a single Rust binary with built-in V8 — no Chromium. Fetch with render-fallback tiers, multi-engine search, screenshots, and persistent login-state sessions (cookies/localStorage export/import), behind HTTP, native MCP (27 tools), and CDP — existing Playwright/Puppeteer code attaches directly. Stealth TLS fingerprints (Chrome/Firefox/Safari/Edge) for risk-controlled sites. Apache-2.0.
Browser Automation auth: None
- 08
lxman/obsbot-mcp
Control an OBSBOT Tiny 2 webcam over USB — pan/tilt/zoom the motorized gimbal in degrees, AI subject tracking, presets, focus/exposure/white-balance, snapshots and recording. Point it at a pixel, or zoom to fit a region of a snapshot it just took. 35 tools (34 on Linux), speaking UVC and the vendor protocol directly with no vendor SDK and no cloud; prebuilt native helpers for Windows, macOS and Linux. npx obsbot-mcp
Embedded System auth: None
- 09
trueoriginlabs/vibatchium
Self-hosted stealth browser automation for unattended agents. Drives real consumer Chrome via Patchright, holding N parallel persistent logged-in profiles on one daemon so sessions survive across runs. Encrypted credential vault with TOTP and IMAP email-code 2FA, a renderer-free TLS-impersonating fetch lane that reuses session cookies, token-frugal page extraction, and prompt-injection scanning on by default.
Browser Automation auth: API Key
- 10
andresolbach/nodriver-mcp-server
Undetected Chrome automation via nodriver, for sites behind Cloudflare, DataDome and similar anti-bot systems. Direct CDP with no ChromeDriver binary, so navigator.webdriver stays undefined. 57 tools whose names match chrome-devtools-mcp, so switching is a config change: accessibility-tree snapshots, input automation, network/console inspection, device emulation, persistent profiles and session save/restore.
Browser Automation auth: None
- 11
yunfeng-enrich/usemymac
Turn a Mac into a remote computer-use MCP server with one command. uvx usemymac up wraps CUA's computer-server (screenshot, mouse, keyboard, windows, apps, accessibility tree) in a per-run token, hides shell and file tools unless opted in, and opens a Cloudflare quick tunnel so Claude Code, Codex, or any MCP client on another machine can use the Mac. Rides an existing CuaDriver daemon when one is running.
OS Automation auth: API Key
- 12
dockndevai/mcp-macos
Safe-by-default control of a Mac: read files, list processes & apps, take screenshots, read the clipboard; and behind explicit opt-ins, run shell commands (allowlisted), AppleScript, move files to Trash, and inject GUI input. Read-only by default with layered access modes, per-path/command allowlists, and human confirmation (MCP elicitation) on the dangerous calls. npx -y @dockndevai/mcp-macos
OS Automation auth: None
- 13
emazaheri/ios-agent
Drive a real iPhone or an iOS Simulator through XCUIAutomation and WebDriverAgent. Screens arrive as a compact digest rather than raw accessibility XML (37,000 tokens to 329 on a long list), element resolution runs server-side through six tiers so a retry costs no model tokens, every action returns the screen it produced, and destructive actions are classified before they run and ask first.
OS Automation auth: API Key
- 14
chryaner/terrarium
Fork, break, and revert real VirtualBox VMs as MCP tools. Forks are linked clones (~0.1s) and revert restores a RAM snapshot in seconds. Drives GUI-only guests through the hypervisor with screenshot, click, scroll, type and keys, so an agent controls a machine with no SSH, guest agent, or network, including Windows XP. Windows host with VirtualBox required. npx -y terrarium-mcp mcp
OS Automation auth: None
- 15
faze79/WPFVisualTreeMcp
#⃣ - Inspect, debug and drive running WPF (.NET desktop) apps: visual tree, dependency properties, data bindings and binding errors, DataContext, and screenshots with open popups/menus included, plus clicking, item selection, text input and keyboard shortcuts. Auto-injects into any running WPF process (x64/x86) with no source changes. dotnet tool install -g WpfVisualTreeMcp
OS Automation auth: None
- 16
maccydee/cute-web-scraper
Web scraping with four escalating fetch tiers (httpx, TLS impersonation, Playwright, stealth) that only escalate on a detected block. Returns clean markdown, or saves large scrapes into SQLite tables you query with read-only SQL to keep them out of the context window. 24 tools covering crawls, products, contacts, PDFs and change tracking. MIT, no API key.
Browser Automation auth: API Key
- 17
snapshot-site/snapshot-site-mcp
Screenshot, visual-diff, and AI page-analysis API for AI agents. Capture any URL as PNG/JPEG/WebP/PDF/HTML, diff two versions of a page to catch visual regressions, and get an AI summary of a page. Hosted remote MCP at https://mcp.snapshot-site.com/mcp with OAuth, or npx -y @snapshot-site/mcp. In the official MCP registry as com.snapshot-site/screenshots.
Browser Automation auth: OAuth
- 18
alexpfau/zigbee2mqtt-mcp
Administer a Zigbee2MQTT estate over its MQTT bridge API — whole-network health report (offline devices, weak links, low batteries, pending OTA, devices rejoining), mesh topology, pairing, binding, reporting intervals and device options. 23 tools across read/safe/destructive tiers; destructive tools are disabled by default. npx zigbee2mqtt-mcp
Home Automation auth: None
- 19
mi60dev/visionaire-engine
Design-debugging context for AI agents from a real Chrome: which CSS rule WINS with file:line, blast radius (how many other elements that rule styles) + a scoped-fix selector, WordPress/Elementor origin attribution, interaction timelines, live CSS fix trials, and pixel/alignment/WCAG-contrast audits. No AI inside — deterministic CDP.
Browser Automation auth: None
- 20
Silbercue/public-browser
Chrome automation over raw CDP — no Playwright, no extension bridge, no single-tab limit. Accessibility-tree refs cached across calls survive rerenders, so a page snapshot costs 1,124 chars against Playwright MCP's 6,084. Server-side plan executor runs multi-step flows in one tool call. Ships a Python API for LLM-free scripting. MIT.
Browser Automation auth: None
- 21
segentic-lab/periscope-mcp
Website & web-app testing built for AI agents rather than raw browser bindings: 66 Playwright tools with hard assertions, auto form-fill, persistent authenticated sessions (incl. interactive login for 2FA/SSO), network mocking, real INP, and accessibility/SEO/GEO (llms.txt, AI-crawler access, WebMCP) + Lighthouse audits.
Browser Automation auth: None
- 22
claymore666/debmatic-mcp
Control a HomeMatic / debmatic CCU (eq-3 home automation) over its JSON-RPC and HM-Script APIs — switch and dim actuators, read sensors, system variables and service messages, run programs, and manage rooms, functions, channel links and device assignments. 25 tools over HTTP or stdio; runs locally against your own CCU.
Home Automation auth: None
- 23
paipaipai666/nexus-browser-mcp
Browser automation via accessibility tree with event-driven deterministic snapshots (MutationObserver quiet-window instead of fixed sleeps), HITL governance gates with redacted JSONL audit, multi-task isolation with self-healing rebuilds, and developer observability (console / JS errors / network / Web Vitals).
Browser Automation auth: None
- 24
tinqiao-oss/clawtouch-mcp
Physical USB HID keyboard/mouse control via a Raspberry Pi Pico 2 running open-source firmware. Exposes move, click, drag, type, key combos, and scroll as MCP tools for any MCP client. Genuine physical HID input on the standard driver path, with a --mock mode for hardware-free trials. pip install clawtouch-mcp
OS Automation auth: None
Snapshot 2026-09-18. Fields shown are what the source registry stated, not a live probe — check the linked documentation before integrating. See the methodology.