Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者提供的说明(英文原文)
Windows screenshots, OCR, and redacted screen recording for AI agents. No desktop app needed.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
链接指向作者自己的文件;我们不托管也不重新打包它。打开文件后,Claude Desktop 会进行安装。
在终端中运行这条命令
浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 dimitar-radenkov 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(6.7.163)于14天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 MCPB。
同一分类中有:Spec Driven Development, Remote Desktop Commander, Signoff
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Five gates that make an agent's 'done' provable: run them, read the verdict, audit the record.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
Pointframe
Let your coding agent test the real Windows UI.
Pointframe lets agents inspect your app, interact with its controls, and run repeatable checks with screenshots and signed evidence.
Free, open source, Windows x64.
Set up your agent | Download the screenshot app
Coding agents: read the complete Windows desktop verification recipe before starting.
pointframe verify run runs your build and test gates and, for each scenario in the spec, launches your built app, drives it through its real controls (invoke, type, check), and writes a verdict with a signed evidence bundle: report.json, evidence/ screenshots and an index.html. A good scenario also carries a negative control, a check with a deliberately wrong expectation that must fail.
30-second demo on the same sample: fail, fix, pass, evidence. Click to watch.
This is a real evidence screenshot from a passing run of the WPF sample (its reference spec is in the repo, so you can reproduce it).
| I want to... | Use | First example |
|---|---|---|
| Give my agent eyes on the desktop, and optionally hands | MCP server | Ask the agent to list your displays; it calls list_displays |
| Script captures, OCR and recordings, or gate an agent on a verified pass | CLI | pointframe displays, then pointframe verify run |
| Capture, annotate and record by hand | Desktop app | Press Print Screen |
The CLI and the MCP server run on their own: neither needs the tray app. Capture, OCR and recording only observe. Interacting with an app's controls is off until you set it up for one app (pointframe verify setup, or a policy file; see desktop testing).
Eleven always-on tools: list_displays, list_windows, capture_monitor, capture_window, read_text_from_monitor, read_text_from_window, search_captures, get_capture, start_recording, stop_recording and get_recording_status. Captures come back inline as an image plus a PNG on disk with SHA-256, monitor geometry and DPI. The opt-in desktop_* tools launch an approved app, click, type and check its UI, and write a signed report. Details: MCP reference.
pointframe displays
pointframe windows pointframe capture --monitor '\\.\DISPLAY1' --output .\shot.png pointframe ocr --monitor '\\.\DISPLAY1' pointframe record --monitor '\\.\DISPLAY1' --seconds 10 --output .\take1.mp4 pointframe verify run
The discovery, capture, OCR and recording commands write a single-line JSON response to standard output, on success and on failure. Exit code 0 is success and 1 a runtime failure with a stable Error.Code. Invalid command-line arguments exit 2 and write usage text to standard error instead of JSON. For other commands, follow their command-specific output contracts in the CLI reference and the verify docs. pointframe mcp serve speaks the MCP stdio protocol. record needs ffmpeg.exe. Every command and flag: CLI reference.
claude --plugin-dir ./plugin/pointframe starts the MCP server and adds the verify-desktop-work skill (plugin README). The plugin is not yet in the plugin directory.Pointframe.Cli-win-x64.zip from the latest release, extract it, and run this from that folder (checksum step: CLI install). Open a new terminal afterwards, so the user Path change takes effect. Then pointframe displays lists your monitors, and the verify guide covers your own app..\Pointframe.Cli.exe install
winget install DimitarRadenkov.Pointframe
Press Print Screen to capture a region, add arrows, text or blur, then copy, save, pin or record. Or take the installer from Releases. Features, settings and shortcuts: desktop app guide.
Does it need the desktop app? No. The CLI and the MCP server are standalone: no tray app and no .NET runtime. The desktop app is the separate screenshot tool for people.
Can it run headless? Build and test gates can, including in CI. Checks of the real UI need an unlocked, interactive Windows session; a Windows service in session 0 cannot capture the user desktop.
What does a pass establish? That the gates and scenarios your spec configures all passed. A spec with only gates passes without launching an app, so UI coverage is only what your scenarios check. Trust a pass only when pointframe verify status reports fresh: true. Freshness covers the Git-visible working tree, the spec and the active-task snapshot. A spec with scenarios but no gates can pass but is never fresh (no_build_gate). Files ignored by .gitignore and verify's own output are not counted. It does not prove your code is correct.
Official release builds of the CLI and the MCP server send anonymous usage counts (command or tool name, outcome, duration bucket, host, MCP client type, version, a random install ID), never screenshots, text or paths. Opt out with POINTFRAME_TELEMETRY_OPTOUT=1 or DO_NOT_TRACK=1; this applies to the CLI and the MCP server only. The desktop app's official builds send the events listed in PRIVACY.md, which holds the full policy; source builds send nothing.
CLI reference | MCP reference | Desktop app guide | Verify docs | Developer guide | Privacy | Contributing | Releases | Website
If you find Pointframe useful, a star on GitHub helps others discover it.
If Pointframe saves you time, consider buying me a beer :
Older links point to these headings. The content now lives in its own file.
Moved to the MCP reference.
Moved to PRIVACY.md.
Moved to PRIVACY.md.