chatgpt-local-workspaceBLOCK
ChatGPT 网页对话直连 Windows 本机:文件读写、精确编辑、命令执行、Git 审阅(OpenAI 官方 Tunnel),桌面程序内嵌实时工作台按对话回放调用时间线。单 EXE、零安装。ChatGPT-driven local Windows workspace over OpenAI Tunnel with an embedded live dashboard; single EXE, zero install.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
简体中文 · English
让 ChatGPT 通过 OpenAI 官方 Tunnel 直接操作你的本机:读写文件、精确编辑、执行命令、查看 Git 变更。桌面程序内嵌实时工作台(WebView2,也可以在浏览器打开),按对话查看每一次工具调用的时间线、命令输出和补丁 diff。单个 EXE,免安装,运行时不需要 Node。
非官方社区项目,与 OpenAI 无隶属关系。
它解决什么:让 ChatGPT 网页对话直接驱动本机的文件、命令与 Git,并用实时工作台逐次回放每个工具调用。codex CLI、@modelcontextprotocol/server-filesystem 跑在终端或 TUI 里;这个插件跑在 ChatGPT 网页里,自带可视化时间线、diff 与命令输出,不用另开终端,也不依赖常驻的 Node 服务。同一个 EXE 也可以作为 stdio MCP 服务器给 Codex 使用(见“Codex 推荐配置”)。
当前版本 2.4.1(2026-10-09,上一公开版本 2.4.0)。各版本变化见 升级说明,接口迁移和上游限制见 2.4.0 发行说明,验证结果见 验证记录。
目录
- 2.4.1 的变化 · 2.4.0 的变化
- 特性一览 · 环境要求 · 快速开始 · 怎么用
- 23 个工具(含 2.4.0 工具删合对照、Codex 推荐配置)
- 实时工作台 · 桌面程序
- 命令执行细节 · 过程可见性 · 版本历史
- 从源码构建 · 更新与替换 · 常见问题 · 安全说明 · 协议兼容性
2.4.1 的变化
- 原始日志折叠 Tunnel 启动噪声:Tunnel Client 0.0.16 在
--log.level info下会把每一步 fx 依赖注入都打成 INFO(每次启动约 240 条),淹没有用的信息。桌面“原始日志”页签现在把这些行折叠为一行“已折叠 N 条 Tunnel 启动内部日志”,每次启动一行。只按精确的事件名折叠(provided、replaced、decorated、supplied、invoking、run、OnStart/OnStop hook executing/executed、initialized custom fxevent.Logger、started),且只折叠 INFO;WARN / ERROR、健
41fa207d9a14OBSERVED · 2026-10-12Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add devspace --env DEVSPACE_OAUTH_OWNER_TOKEN=${DEVSPACE_OAUTH_OWNER_TOKEN} --env WORKSPACE_TUNNEL_SMOKE_KEY=${WORKSPACE_TUNNEL_SMOKE_KEY} -- npx -y @waishnav/[email protected]{
"mcpServers": {
"devspace": {
"command": "npx",
"args": [
"-y",
"@waishnav/[email protected]"
],
"env": {
"DEVSPACE_OAUTH_OWNER_TOKEN": "${DEVSPACE_OAUTH_OWNER_TOKEN}",
"WORKSPACE_TUNNEL_SMOKE_KEY": "${WORKSPACE_TUNNEL_SMOKE_KEY}"
}
}
}
}Exposed tools (8)
5 read · 3 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
apply_patch | write | |
claude | read | A profile that shadows the raw provider. |
claude-reviewer | read | Unavailable profile. |
custom | read | Use a custom model. |
exec_command | write | |
reviewer | read | Review changes. |
tdd | read | Tests first |
write_stdin | write |
Trust audit
BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | FAIL |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (1 observation(s))
- Network
- declared (5 observation(s))
- Shell
- declared (4 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- none-found
Findings (13)
review-payload.tsx
join(home, ".netrc"),
Microsoft.Web.WebView2.Core.dll
Microsoft.Web.WebView2.WinForms.dll
WebView2Loader.dll
name: "../../generated.png",
- `node scripts/dashboard-preview.cjs http://127.0.0.1:<port>/`: observe a running old instance read-only.
- `node scripts/dashboard-preview.cjs http://127.0.0.1:<原端口>/`:只读观察运行中的旧实例。
- 正式真实会话 https://chatgpt.com/c/6aaa5998-98c8-83e8-8305-59a5070d7248 成功登记“独立工作台正式验收”,随后显式绑定该实际 chat ID,并调用带相同 thread_id 的 file_info。正式本地网页自动出现对应名称、ChatGPT 跳转链接及 README.md 操作时间线(2 ms)与回执。当前本机地址 http://1
- 新版页面 http://127.0.0.1:22775/ 只读桥接 http://127.0.0.1:36637/。Host/Origin/Fetch-Site/GET 与路由限制已验证。旧桌面按钮仍指向原页面;dist-next 编译完成,未覆盖正式程序。
- 复核(19:12–19:15,样例页 http://127.0.0.1:3203/):点击 WorkspaceDetail.cs 读取显示第 96–117 行正文,点击 local-workspace.ico 显示“不是文本”且无正文;npm test 36/36 通过(16.3 s)。
cn, radix-ui, class-variance-authority, esbuild, lucide-react, react, react-dom, shadcn
@agentclientprotocol/sdk, @clack/prompts, @earendil-works/pi-coding-agent, @modelcontextprotocol/ext-apps, @modelcontextprotocol/node, @modelcontextprotocol/sdk, @modelcontextprotocol/server, @pierre/
Gates applied: no_behavioural_pass.
41fa207d9a14full audit observations/trust-audit/mcp-server/csl19980820__chatgpt-local-workspace.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-12 | 41fa207d9a14 | BLOCK | D | 69 | first audit |
Questions
What is the chatgpt-local-workspace MCP server?
ChatGPT 网页对话直连 Windows 本机:文件读写、精确编辑、命令执行、Git 审阅(OpenAI 官方 Tunnel),桌面程序内嵌实时工作台按对话回放调用时间线。单 EXE、零安装。ChatGPT-driven local Windows workspace over OpenAI Tunnel with an embedded live dashboard; single EXE, zero install.
What tools does chatgpt-local-workspace expose?
8 in total: 5 read-only, 3 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is chatgpt-local-workspace safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (69/100) and found 2 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What credentials does chatgpt-local-workspace need?
It reads DEVSPACE_OAUTH_OWNER_TOKEN and WORKSPACE_TUNNEL_SMOKE_KEY from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How current is this page?
The grade is for one exact copy of the source (41fa207d9a14), read on 2026-10-12. The repository is watched and re-audited when it changes.