Terminator MCP Agent

What louis030195/terminator-mcp-agent does

The louis030195/terminator-mcp-agent MCP server connects MCP-compatible AI clients to Windows desktop automation. It can locate interface elements using metadata such as names, roles, windows, control types, and automation identifiers, then perform actions including clicking and typing. It also supports launching and managing applications, inspecting active or open windows, controlling browsers through a Chrome extension, and handling multiple monitors.

This places the project in the 🖥️ OS Automation category. It is suited to tasks that cross application boundaries, such as creating a cloud instance and opening a CLI connection, reviewing service logs, or exercising application features after code changes. The README also describes workflow recording for turning human interactions into deterministic automation.

How it works

The server is started as a standard MCP command and communicates with an AI client over the client’s configured process connection. The documented command uses npx -y terminator-mcp-agent@latest, so Node.js package execution is part of the setup path.

Automation relies on Windows accessibility information and element selectors rather than requiring image matching for every interaction. To build selectors, inspect an application with Accessibility Insights for Windows or Inspect.exe. These tools expose metadata such as an element’s name, role, control type, and AutomationId; that information can then be used in selectors such as a name, role, or window selector.

The project also describes browser-session reuse, including existing cookies and authentication, and operation that does not take over the user’s cursor or keyboard. The supported implementation is Windows-only, including the MCP agent itself.

Setup and configuration

For Claude Code, add the server with:

claude mcp add terminator "npx -y terminator-mcp-agent@latest"

Other clients can add an MCP entry with npx as the command and -y, terminator-mcp-agent@latest as its arguments. The example configuration sets LOG_LEVEL to info and RUST_BACKTRACE to 1; these variables are useful for controlling logging and Rust backtraces when troubleshooting.

Cursor and VS Code provide documented install links in the repository, and the README also names Windsurf and Claude among supported client examples. The louis030195/terminator-mcp-agent MCP server requires a Windows environment for its documented automation features. macOS and Linux installation or automation support is not available.

Tools and capabilities

• Find Windows UI elements by accessibility metadata.
• Click and type into supported interface elements.
• Launch, list, and manage applications.
• Inspect active and open windows.
• Automate browsers through the Chrome extension.
• Record workflows for deterministic replay.
• Work with multiple displays.
• Capture displays or individual elements.

The project lists TypeScript, workflow, CLI, key-value, Python, and Rust libraries separately. Those libraries are not required to use the MCP server, and the README marks Python support as partial while listing the other Windows support as available.

Limitations and notes

The louis030195/terminator-mcp-agent MCP server currently supports Windows only. Its feature table marks macOS and Linux as unsupported for the MCP agent, core automation, browser automation, recording, monitor management, and capture. Browser control depends on the Chrome extension. Reliable selectors require access to the Windows accessibility tree, so applications that do not expose useful accessibility metadata may need additional investigation.

The repository describes the project as MIT-licensed. For detailed troubleshooting and server logs, it directs users to contact the project through the linked Mediar site.