Progress Watch
Report progress from any script or agent and get notified when it finishes.
사용해야 할까요
품질 및 안전성
도구 정의와 프로토콜 준수에 대한 자동 분석을 기반으로 합니다.
컨텍스트 비용
이는 서버의 도구가 모델의 컨텍스트에 로드될 때마다 소비되는 대략적인 토큰 수입니다. 수치가 높을수록 다른 작업에 사용할 수 있는 주의가 줄어듭니다.
설치
원클릭 설치
`claude_desktop_config.json` 파일에 다음을 추가하세요:
{
"mcpServers": {
"progresswatch": {
"url": "https://progress.watch/mcp/{space_uuid}"
}
}
}원격 엔드포인트
https://progress.watch/mcp/{space_uuid}streamable-http할 수 있는 일
도구 목록
도구 (4)
🟡create_space(title, icon)
Create a space — the container tasks live in, and what the user opens to watch them. Only call this when no space is configured for this connection and creating a task failed for that reason. If a space is already configured, use it: making a second one splits the user's tasks across two screens they have to add separately. Returns the space uuid and the URL to watch it. Give the user that URL — until they open it, nothing you report is visible to them, and the uuid is the only way back to it.
입력 스키마
{
"type": "object",
"properties": {
"title": {
"type": "string",
"maxLength": 100,
"description": "What this collection of tasks is for, e.g. \"Nightly builds\". At most 100 characters; longer is cut to 100, the last one an ellipsis."
},
"icon": {
"type": "string",
"description": "One character shown beside the name. An emoji reads best, e.g. \"🌙\"."
}
},
"required": [
"title"
]
}🟡create_task(title, parent_uuid, space_uuid, source)
Create a task the user can watch from anywhere while it runs. Call this first in every turn you do any work, reading and answering included, without being asked and however small the work looks: the user connected this server to watch exactly that work, and a turn that reports nothing looks to them like a turn that never happened. Returns a task uuid to report against. If you decompose work into steps, create one child task per step by passing parent_uuid. Nesting is one level only: a child cannot have children. A parent's progress is averaged from its children automatically, so never report progress on a parent yourself. Create the whole tree before starting, then report each step as it happens and complete it when it is actually done — not all of them at the end. A board that stays empty for the whole job and turns green at the finish is worth nothing to whoever is watching it. Keep creating steps as work appears. The tree is a live picture, not a plan you fixed at the start: something you did not foresee gets its own step when you find it, rather than being folded into whichever step is already open. A parent whose bar drops because it gained a step is telling the truth.
입력 스키마
{
"type": "object",
"properties": {
"title": {
"type": "string",
"maxLength": 200,
"description": "What this task is doing, in a few words. At most 200 characters; longer is cut to 200, the last one an ellipsis."
},
"parent_uuid": {
"type": "string",
"description": "Make this a step of an existing task. One level only."
},
"space_uuid": {
"type": "string",
"description": "Only needed to override the space configured for this connection."
},
"source": {
"type": "string",
"maxLength": 100,
"description": "What is reporting, e.g. the agent or tool name. At most 100 characters; longer is cut to 100, the last one an ellipsis."
}
},
"required": [
"title"
]
}🟡update_task(task_uuid, current, end, values)
Report progress on a task. Call periodically — every meaningful step, not every loop iteration. Call it once when the work actually begins, too. A task that has been created but never reported reads as "waiting for data", which nobody can tell apart from a reporter that died before its first write. If the step has nothing to count, send current 0 and end 1: that reads as running, and it is honest. `current` and `end` are raw counts, not a percentage: the user sees "1200 of 50000 pages", which tells them something "2.4%" does not. `end` may change as you learn more; send the new number and the bar recalculates. Every call replaces the whole state. Send everything each time — anything you leave out is cleared, not kept. Put a one-line status in values.log; it holds the last line only, not a history. A task uuid is all this needs — it already knows which space it belongs to.
입력 스키마
{
"type": "object",
"properties": {
"task_uuid": {
"type": "string",
"description": "The uuid returned by create_task."
},
"current": {
"type": "number",
"description": "How much is done so far."
},
"end": {
"type": "number",
"description": "How much there is in total."
},
"values": {
"type": "object",
"maxProperties": 20,
"description": "Flat object of extra numbers or strings, e.g. {\"errors\": 3, \"log\": \"retrying /foo\"}. At most 20 keys of at most 40 characters, or the request is refused with 400. A string is cut to 1000 characters, the last one an ellipsis."
}
},
"required": [
"task_uuid"
]
}🟢complete_task(task_uuid, values)
Mark a task finished. This is what sends the user's push notification, so call it when the work is actually done — including when it failed, with the failure recorded in values. Call it as part of the step that finishes the work, not in a sweep once everything is done. Closing a whole tree at the end is cheaper for you and leaves the user watching an empty board for the entire job. Nothing finishes a parent for you: complete it once its last step is done. Completing a parent also stops every step under it that is still open, so a job that ends early is closed by completing the parent alone. Those steps show as stopped, not done, with their numbers as they were.
입력 스키마
{
"type": "object",
"properties": {
"task_uuid": {
"type": "string",
"description": "The uuid returned by create_task."
},
"values": {
"type": "object",
"maxProperties": 20,
"description": "Final numbers or a closing log line, e.g. {\"status\": \"failed\", \"log\": \"exit 1\"}. Same limits as update_task."
}
},
"required": [
"task_uuid"
]
}커뮤니티
증거