Control your desktop and execute commands from AI agents

Desktop Commander MCP gives an AI agent full access to a local machine: executing terminal commands and interacting with long-running processes or SSH sessions, reading and writing files across formats (plain text, Excel, PDF, DOCX) with surgical/targeted editing rather than full-file rewrites, listing/killing/interacting with running processes, and running code (Python, Node.js, R) directly in memory without needing to save it to disk first. It also handles instant ad-hoc data analysis of CSV, JSON, and Excel files on demand, and supports Remote MCP so it can be controlled from ChatGPT or Claude's web interface rather than only a local desktop client.

It works with Claude Desktop, Cursor, VS Code, Windsurf, and any other MCP-compatible client, making it a general-purpose local-machine bridge rather than a tool tied to one specific AI product.

Use this connector when an AI agent needs genuine local-machine capability beyond a sandboxed code interpreter - running real terminal commands, managing actual OS processes, editing files across multiple formats with precise in-place changes, or executing code in memory for quick iteration without file management overhead. It's particularly suited to development workflows, data analysis tasks, and system administration where the agent needs direct machine access rather than a restricted execution environment.

Because it grants full terminal, filesystem, and process access, it should be installed and used with the same care as giving any tool shell-level access to a machine - it is not a sandboxed or restricted execution environment, and its capabilities (killing processes, writing arbitrary files, running arbitrary code) carry the same risk profile as direct terminal access for whoever's machine it's connected to.

Input is a natural-language request routed through an MCP-compatible AI client - a command to run, a file to read/edit, a process to inspect or kill, or code to execute in memory. Output is the command's terminal output, the file's read/edited content, process status, execution results from in-memory code, or an ad-hoc data analysis result.

npx @wonderwhy-er/desktop-commander@latest setup

• Terminal command execution, including long-running processes and SSH sessions
• Full filesystem read/write across text, Excel, PDF, and DOCX formats with surgical editing
• Process management: list, kill, and interact with running processes
• In-memory code execution (Python, Node.js, R) without saving files
• Instant analysis of CSV, JSON, and Excel data on demand
• Remote control via Remote MCP from ChatGPT and Claude web

Run npx @wonderwhy-er/desktop-commander@latest setup for Claude Desktop setup. It's also available via GitHub (wonderwhy-er/DesktopCommanderMCP), the npm package @wonderwhy-er/desktop-commander, and configurable in any other MCP-compatible client (Cursor, VS Code, Windsurf), with a Discord community for support.

Developers and power users who want their AI agent to have genuine local-machine capability - running commands, managing files and processes, executing code in memory - across whichever MCP-compatible client they use, rather than being limited to a sandboxed execution environment.