
Pir Mcp
io.github.Solteris-Dev由 GitHub 账号 Solteris-Dev 发布v0.1.0更新于 Oct 24, 2026
Motion sensor for AI agents: tells when a screen region changes or goes still. Scores, never pixels.
概览
让助手监视指定的屏幕区域,以数值分数而非图像报告区域是否发生变化或已静止。
- 可以做什么
- 它定义具名的屏幕矩形或窗口并按需采样,在粗粒度平均颜色网格上返回两个变化分数:rmse 表示整个区域是否移动,peak 表示变化最大的单个格子。工具包括由人绘制区域的 pir_pick_region 和 pir_pick_mask、用于脚本化几何的 pir_define_region 和 pir_define_window、一次性检查的 pir_sample,以及阻塞等待区域变化或静止的 pir_wait_for_change 和 pir_wait_for_stillness。它从不返回像素、图像文件或任何可还原为图片的内容。
- 适用场景
- 适合驱动桌面的助手在不看屏幕的情况下判断某处是否发生变化或停止变化,例如等待构建输出更新,或检测状态栏是否卡死。适用于长时间监视,此时截图既无必要也过于暴露。
- 运行要求
- 以 npm 包(npx pir-mcp)在本地运行,需要 Node 20+。需要能向标准输出写入二进制 PPM 的截图工具,通过 PIR_CAPTURE_CMD 配置(wlroots 下默认 grim)。可选的选择器与窗口命令通过 PIR_SELECT_CMD、PIR_PICK_WINDOW_CMD 和 PIR_WINDOW_GEOMETRY_CMD 设置;留空会禁用相应工具。仅限桌面。
安装
在 SourceWeft 中
- 打开 控制台中的 Pir Mcp,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
pir-mcp
A motion sensor for AI agents.
"Gives your LLM agents enough info to be a helpful "watcher" without being a privacy nightmare like Windows Recall" -Solteris-Dev
[pir-mcp MCP server – quality and maintenance score on Glama]
A PIR sensor tells you that something moved, never what. This is the same idea for a screen: an MCP server that watches named rectangles and answers "did this change?" and "has this gone still?" with a number, a size and a timestamp. No tool in it returns pixels, image files, or anything that could be turned back into a picture.
It exists because an agent driving a desktop keeps needing the same two things: wake me when the build output moves and tell me if this status bar has stopped ticking. Both are motion questions. Neither needs a screenshot.
What it does
pir_pick_regionask the person at the screen to draw the rectangle, the way a regional screenshot works (slurp, slop). The human chooses what is sensed; the agent gets the geometry. Use this whenever someone is there.pir_pick_maskthe same gesture for a patch inside a region to ignore (a clock, a spinner, a caret).pir_define_regionthe scriptable path: name a rectangle by global layout coordinates, with optional masks.pir_pick_window/pir_define_windowwatch a window instead of a fixed rectangle. Its geometry is looked up again before every capture, so the region follows the window when it moves; a resize counts as maximal change. The pick offers your visible windows as boxes to click.pir_samplecapture once, score against the previous sample of that region. Cheap; also the way to check the capture command works.pir_wait_for_changeblock until the region departs from how it looked when the call began, or time out. A doorbell.pir_wait_for_stillnessblock until nothing has changed forstill_for_ms, or time out. A short window means "the animation settled, safe to act". A long window on something that should keep changing means it is frozen, andstill=trueis the alarm.pir_list_regions,pir_remove_regionhousekeeping.
Every comparison reports two numbers over a coarse grid of averaged colour cells (at most 64 across, each at least 4 px), both 0 for identical and 1 for black against white:
rmseover the whole grid: did the region as a whole move? A cursor edge or an antialiased caret scores about 0.02; a dialog opening 0.1 or more. This is the default metric, threshold 0.05.peakthe single most-changed cell: did anything in it move? A clock digit flipping inside a 1920-wide status bar scores rmse 0.018 (invisible to the default) but peak 0.145. Usemetric=peakfor a small thing that should tick inside a larger region.
Averaging is what makes this a sensor rather than a camera: 64 cells cannot be read.
What it deliberately does not do
- It never returns image data. There is no snapshot tool and none is planned. If an agent needs to see, use the screenshot tool your environment already has, so that choice stays explicit and yours.
- It never writes frames to disk. A capture lives in memory for the milliseconds it takes to reduce it to a grid, and only the grid of the last sample is kept per region.
- It does not capture on its own. Every sample is a tool call the agent made,
visible in the transcript, and
pir_pick_regioncarries apurposeline the agent has to write down before you draw.
What it does reveal, so you can decide whether that is acceptable: that a given rectangle changed, at a given time, by a given magnitude. On a region covering a chat window that is presence information. Choose regions accordingly; the server has no opinion.
The capture itself is delegated to a command you configure, run with your
privileges. The default is grim, so nothing here has screen access that you
did not already give to grim.
Install
It is on npm as pir-mcp and in
the official MCP Registry
as io.github.Solteris-Dev/pir-mcp, so most clients can run it with
npx -y pir-mcp and no clone:
From source:
Requires Node 20+ and a screenshot tool that can write binary PPM to stdout.
Capture command
PIR_CAPTURE_CMD is a template; {x} {y} {w} {h} are substituted and the
result runs under sh -c. It must print a binary PPM (P6) to stdout.
Coordinates are whatever the capture tool uses. On a wlroots layout that is the global layout, so an output placed left of the primary has negative x.
Selector command
PIR_SELECT_CMD runs when an agent calls pir_pick_region or
pir_pick_mask and must print x,y wxh. Set it to an empty string to remove
those tools entirely.
Window commands
PIR_WINDOW_GEOMETRY_CMD gets {id} substituted and must print x,y wxh
for that window; PIR_PICK_WINDOW_CMD must print the id of the window the
person chose. Both default to Hyprland (via jq and slurp -r). Set either
to an empty string to remove the corresponding tools. Window ids are limited
to [A-Za-z0-9_.:-] before they reach a shell.
Only the Hyprland pair has been run; the others are written from the tools' documentation.
Claude Code
Blocking calls default to 55 s and are capped by PIR_MAX_WAIT_MS (540 s).
Keep the cap under your host's MCP tool timeout; an agent that needs to watch
for longer just calls again.
Other settings
A regions file looks like:
Example
The case that produced this: a status bar that occasionally froze for hours while its process looked healthy. The clock in it should change every minute, so a bar that is still for three minutes is a frozen bar.
still=true comes back only if the bar stopped; otherwise the call returns
still=false at the timeout and the agent calls again. Nothing on the screen
was ever seen, and the person drew the rectangle themselves.
Contributing starting points
contrib/macos/ holds an AppKit rectangle selector and capture recipe,
written blind and untested. If you run it on a real Mac, fix what breaks and
send it back.
Tests
License
MIT.
来源:README.md,提交 833ded4
工具
0版本历史
1- v0.1.0最新Sep 17, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分