Terminator MCP Agent

Official

Computer-use MCP server that automates Windows desktop and browser apps via the accessibility tree.

Unverified

stdio (local)

No auth

Rust

View repo 1.5k Website npm

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run

npx -y terminator-mcp-agent@latest --add-to-app

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "terminator-mcp-agent": {
      "command": "npx",
      "args": [
        "-y",
        "terminator-mcp-agent@latest"
      ],
      "env": {
        "LOG_LEVEL": "info",
        "RUST_BACKTRACE": "1"
      }
    }
  }
}

Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf

Before you start

Windows (Windows 10/11 or Windows Server 2016+); macOS and Linux are not supported
Node.js v16+ (for the npx-based launch)
An MCP-compatible client (Claude Code, Cursor, VS Code, Windsurf, etc.)
Chrome extension installed for browser automation features
Accessibility Insights for Windows or Inspect.exe (optional, to author reliable selectors)

About Terminator MCP Agent

A computer-use MCP server from mediar-ai that controls an entire Windows desktop, letting LLMs and agentic clients automate tasks across any application and the browser through the Windows Accessibility tree without relying on vision models or screenshots. It is built on the Rust-based Terminator library and ships as an npm package (terminator-mcp-agent) for one-command install into Claude Code, Cursor, VS Code, and other MCP clients. Beyond single actions, it provides an execute_sequence tool to run deterministic, parameterized YAML/JSON workflows with selectors, conditional jumps, state persistence, and embedded JavaScript/Python—calling AI only when recovery is needed.

Tools & capabilities (24)

get_window_tree

Retrieve the accessibility/UI element tree for a window or process, with options to limit depth, start from a selector, and choose output format.

get_applications_and_windows_list

List running applications and open windows so workflows can discover and target them.

open_application

Launch an application by path or name (e.g. calc.exe).

click_element

Click a UI element resolved by a selector, with actionability validation (visibility, enabled, stable).

invoke_element

Invoke an element via the accessibility API as an alternative to a physical click.

activate_element

Bring an element or its window to the foreground / give it focus.

type_into_element

Type text into a targeted UI element.

press_key

Send keyboard key presses to a focused element.

scroll_element

Scroll a UI element in a given direction.

select_option

Select an option in a dropdown or combo box element.

set_value

Set the value of an input or editable element directly.

mouse_drag

Perform a mouse drag from one point/element to another.

validate_element

Check whether an element matching a selector exists / is in an expected state (used for conditional branching).

wait_for_element

Wait until an element meets a condition (e.g. exists) within a timeout, optionally capturing the UI tree.

navigate_browser

Navigate the browser to a URL using the Chrome extension integration.

execute_browser_script

Run JavaScript in the browser/page DOM and return results, with variable injection support.

capture_screenshot

Capture a screenshot of a display, window, or element.

run_command

Run a shell command, or execute JavaScript/Python (engine mode) with access to desktop automation APIs and automatic data passing between steps.

execute_sequence

Run a full deterministic workflow of steps with variables, inputs, selectors, groups, conditional jumps, partial execution, state persistence, and output parsing.

read_file

Read a file from disk for use within workflows.

write_file

Write content to a file on disk.

edit_file

Edit an existing file on disk.

glob_files

Find files matching a glob pattern.

grep_files

Search file contents with a pattern.

When to use it

Automate repetitive tasks across any Windows desktop application without legacy RPA complexity
Drive browser workflows using the user's existing logged-in session and cookies
Record human workflows and generate deterministic YAML automations that run at CPU speed
Run automation on headless cloud VMs (Azure/AWS/GCP) via virtual display support for scalable automation farms
Spin up infrastructure or inspect logs through desktop/CLI apps (e.g. create a GCP instance, check Vercel error logs)
Test an app's new features end-to-end based on recent commits

Security notes

Grants AI assistants full control over the desktop and any application, including the user's logged-in browser session, cookies, and auth. Never hard-code credentials in workflows—use environment variables or a secret manager. Workflows are code: store them in version control and review changes. Run with elevated/administrator privileges only when accessibility features require it.

Terminator MCP Agent FAQ

Which operating systems are supported?

Windows only. The README explicitly states macOS and Linux are not supported.

How do I install it in Claude Code?

Run the one-liner: claude mcp add terminator "npx -y terminator-mcp-agent@latest". For other clients, add the mcpServers block with command "npx" and args ["-y", "terminator-mcp-agent@latest"], or run npx -y terminator-mcp-agent@latest --add-to-app.

Does it take over my mouse and keyboard?

No. It runs in the background using the accessibility tree and your browser session, so it does not hijack your cursor or keyboard or require you to re-login.

Can it run over HTTP instead of stdio?

Yes. Launch with -t http to expose POST /mcp plus /health, /status, and an SSE /events endpoint for realtime workflow events. Concurrency is controlled via the MCP_MAX_CONCURRENT environment variable (default 1).

How does it stay reliable and fast?

Terminator pre-trains workflows as deterministic code and only calls AI when recovery is needed, claiming greater than 95% success rate and much faster execution than general computer-use agents.

#computer-use #desktop-automation #windows #browser-automation #accessibility #rpa #workflow #rust

Alternatives to Terminator MCP Agent

Compare all alternatives →

Playwright MCP Server

Developer Tools

24k

Microsoft's official browser-automation MCP using Playwright's accessibility tree (no vision model).

Featured

Verified

stdio (local)

No auth

TypeScript

12 tools

Updated 13 days agoRepo

Context7 MCP Server

Developer Tools

28k

Up-to-date, version-specific library documentation injected into your coding agent.

Verified

stdio (local)

API key

TypeScript

2 tools

Updated 17 days agoRepo

Serena

Developer Tools

12k

LSP-powered coding agent toolkit: semantic symbol search, references and structural edits.

Verified

stdio (local)

No auth

Python

11 tools

Updated 15 days agoRepo

Compare Terminator MCP Agent with:

vs Playwright MCP Server vs Context7 MCP Server vs Serena vs Mobile MCP