"A guide to implementing web browser control functionality" matching MCP servers:

Enables Codex to manage a local Firefox session by listing, grouping, opening, activating, navigating, and closing non-private tabs and native tab groups through a secure local native-messaging bridge.

Deterministic, LLM-first control of Word documents (.docx). An agent reads the outline, finds text by phrase, and applies edits as tracked changes and comments in one atomic batch, with a safety layer that refuses edits based on a stale read and rolls back failed batches. It works on .docx files directly, and can also drive an open document in Microsoft Word on Windows or in LibreOffice Writer.

Enables AI agents to drive a live AutoCAD session over stdio with transactional, serialized writes: inspecting and saving documents, managing layers, creating and editing line/circle/arc/polyline/rectangle/text entities, querying selections, setting checkpoints, and exporting DXF. Every request is validated against strict schemas with document revision checks and save-reopen geometry readback, and a headless backend is available when no AutoCAD instance is running.

Enables AI agents to drive a connected Android phone over ADB with 28 tools covering screenshots and UI inspection, taps/swipes/typing, app install and launch, file transfer, wireless pairing, and one-call scripted flows like tap-by-text and wait-for-text.

Lets an MCP client see and control apps on a local X11 Linux desktop by clicking, typing, scrolling, dragging, and pasting text, with a visible Stop control that returns input to the user.

MCP server that provides computer control capabilities including mouse movements, keyboard actions, screenshot capture with OCR, and window management through a unified API.

Enables AI agents to operate websites and Windows desktop applications by reading and acting on webpage elements, exposed UI controls, and screenshots with mouse and keyboard input. Combines browser, desktop, and computer-control channels in a single task, with an optional target selector for batched actions.

A native MCP server that gives AI assistants full Chrome control via the DevTools Protocol and an unpacked extension, including semantic element refs, real input, screenshots, and browser/tab/window management. It exposes 43 callable tools with a visible control overlay and an optional read-only security profile.

MCP server for AI to operate Windows apps via screen OCR and background input, featuring structured WhatsApp UI tools for reading chats, detecting unread messages, and sending messages without stealing focus.

Enables GUI automation by capturing screen, controlling mouse/keyboard, and targeting controls via accessibility layer; optionally supports voice commands and synthesis.

An MCP server for YouTube automation using Playwright and Gemini AI agent. Enables browser control for YouTube operations like searching, playing videos, skipping ads, and managing playback through natural language commands.

Local MCP server for A-share stock trading via Tonghuashun, offering account/position queries, buy/sell/cancel orders with risk controls and forced user confirmation; currently simulated with a reserved interface for real broker channels.

Enables agents to convert Markdown into WhatsApp-compatible formatting, including styled text, lists, and tables that fit a phone's monospace width.

Bridges the editing capabilities of an already-open DingTalk document in the browser to any MCP host via standard stdio, forwarding tools to the page's internal MCP server over a local WebSocket. Supports read/write operations through suggested edits, pairing-based security, and zero runtime dependencies.

Enables MCP clients to turn Make blueprints and Zapier exports into n8n workflow files, with a per-step report of what converted, what to review, and what must be finished by hand. Also lists zaps, explains how individual steps are handled, checks workflow files for broken links or placeholders, and can upload the result to an n8n instance as an inactive workflow.

A local stdio MCP service that lets an authorized AI assistant see and operate one Android phone through ADB, with screenshots, on-device OCR, text/coordinate taps, swipes, and app launching, plus optional scrcpy-based Chinese and emoji input. Includes zoomed inspection, a 0-1000 coordinate ruler, and crosshair preview with post-tap feedback so actions can be verified against the real screen.

An MCP server that enables AI assistants to control TVs on a local network through natural language commands. It currently supports Roku devices, allowing users to launch apps, manage playback, and navigate menus.

Enables Claude Code to read and operate macOS app interfaces through the accessibility tree, pressing buttons, setting text fields, typing with IME support, sending shortcuts, scrolling, dragging, and capturing screenshots. It drives background apps without moving the mouse or stealing focus from the user's current work.

Enables AI agents to observe and operate a real Windows or Linux desktop like a human — snapshotting screen geometry, active window, OCR text and normalized UI elements, then clicking, typing, scrolling and managing windows with built-in verification and recovery. Backends are chosen automatically per action (UIA/AT-SPI → OCR → vision → PyAutoGUI), so browsers, IDEs, office suites and canvas apps are all controllable without any browser-automation MCP.

Enables AI assistants to inspect and safely control OBS Studio over obs-websocket v5, exposing scenes, inputs, filters, outputs and video settings as read and write tools that default to dry runs and refuse changes while live unless explicitly confirmed. It also provides Source Record and IRL readiness checks, a BRB auto-switcher, and a scene timeline log for cutting clips from VODs.