Windows Local MCP
Allows ChatGPT (OpenAI) to access local files, write files, view the screen, and control the Windows desktop via MCP.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Windows Local MCPTake a screenshot and describe what's on my screen."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Windows Local MCP
中文: 让 ChatGPT 在 Chat 模式中通过 MCP 读取和写入本机文件、查看并操作 Windows 桌面,并在本机明确授权后直接运行 Git、构建、测试等命令,无需依赖终端窗口焦点。命令以当前 Windows 用户权限运行,不自动提权,也不是操作系统沙箱。
English: Give ChatGPT in Chat mode MCP-based access to local file reads/writes and Windows desktop control, plus opt-in direct execution of Git, build, test, and other local commands without terminal focus. Commands run with the current Windows user's permissions, do not elevate automatically, and are not an OS sandbox.
中文完整说明:README.zh-CN.md
Full English documentation: README.en.md
安全说明 / Security: SECURITY.zh-CN.md · SECURITY.en.md
验证记录 / Validation: VALIDATION.zh-CN.md · VALIDATION.en.md
License: MIT
权限与风险 / Permissions and risk
中文: 这是高权限本机工具,不是操作系统沙箱。启用后,模型可在当前 Windows 用户本来拥有的权限范围内读取/修改文件,并在获得桌面工具权限后向普通应用发送鼠标和键盘输入。目标锁、一次性截图编号、暂停、备份和审计只能降低误操作风险,不能把模型变成低权限进程。处理密码管理器、支付、敏感账号或重要生产数据时应暂停服务;不要把 Tunnel 凭据、私密截图或本机状态提交到仓库。
English: This is a high-privilege local tool, not an operating-system sandbox. A connected model can read or modify files within the current Windows user's permissions and can send mouse/keyboard input to ordinary applications through the desktop tools. Target locking, one-use screenshot IDs, pause controls, backups, and audit logs reduce mistakes but do not create a low-privilege security boundary. Pause the service around password managers, payments, sensitive accounts, or important production data, and never commit Tunnel credentials, private screenshots, or local state.
See SECURITY.zh-CN.md / SECURITY.en.md for details.
Related MCP server: MCP ChatGPT Full PC Dev
Local command execution / 本机命令执行
Version 0.2.0 adds opt-in command_start, command_poll, and command_cancel tools. They do not require desktop focus and do not elevate privileges. Commands have current-user access, not sandbox isolation. Enable only through local operator controls; see command tools and limits.
0.2.0 新增默认禁用的本机命令工具。开启后可直接运行 Git、构建和测试,不必保持终端前台;不自动提权,也不是系统沙箱。只能在本机确认开启,运行中的 MCP 不得自行修改许可或安装目录。详见使用方法、限制与升级要求。
File editing and search / 文件编辑与搜索
0.3.0 新增 edit_text_file、search_files 和 search_text,支持带版本检查的精确局部编辑、递归文件名搜索和有界文本搜索。直接使用文件接口,不要求开启命令执行。用法、编码和边界见文件工具说明。
Version 0.3.0 adds exact partial edits with read-associated versions, recursive filename search, and bounded literal text search. Command opt-in is not required. See file tools, encodings and limits.
This server cannot be deployed
Maintenance
Related MCP Connectors
Use your Mac, Windows or Linux computer from ChatGPT, Claude or Codex: files, commands, documents.
MCP connector that lets ChatGPT list, search, and run your Apple Shortcuts via a local Mac agent
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Authenticated MCP Agent (Openai)
Related MCP Servers
- AlicenseNot gradedqualityFmaintenanceEnables ChatGPT/Codex to read, search, and edit files in a single allowed folder on Windows through OpenAI Secure MCP Tunnel.ISC
- FlicenseCqualityDmaintenanceEnables ChatGPT to control a Windows PC remotely via OpenAI Secure MCP Tunnel, executing file operations, PowerShell commands, and system actions through a local MCP server.15-
- AlicenseNot gradedqualityAmaintenanceEnables ChatGPT to access approved local Windows folders, run commands, control the desktop, and manage session history through a permission-bounded MCP server.4,416MIT
- AlicenseNot gradedqualityAmaintenanceEnables ChatGPT on Windows to securely access codebases, Git, terminals, language servers, debuggers, SQLite, local HTTP services, adaptive project memory, engineering skills, checkpoints, audit logs, and sandboxed command execution through MCP tools.MIT