Обнаруживайте сбои cron, о которых молчит ваш мониторинг. MCP-сервер, который предоставляет данные о состоянии запланированных заданий — запуски, просроченные задания и «тихие сбои», которые завершаются с кодом 0, но не дают полезного результата — любому агенту Claude или клиенту с поддержкой MCP. Работает «из коробки» с планировщиками OpenClaw, системным cron и таймерами systemd.
Каждая команда, запускающая запланированные задания, сталкивалась хотя бы с одним из этих случаев:
Тихий сбой — задание выполнилось, вернуло код выхода 0, но не произвело полезного вывода (cron для веб-поиска вернул пустоту, бэкап записал файл размером 0 байт, дайджест-рассылка отправилась с <no rows> в теле письма). Традиционный мониторинг видит «зеленую галочку», но данные все равно повреждены.
Просрочка без оповещения — задание не запускалось 3 дня; никто не заметил, потому что никто не следил.
Дрейф последнего успешного запуска — задание запускается каждый час, но за последние 12 попыток успешно выполнилось лишь однажды; все считают, что оно в порядке, так как последний запуск был «зеленым».
Разрыв в журнале аудита — вам нужно знать, когда конкретное задание завершилось в последний раз для проверки соответствия требованиям, а единственный «лог» — это вывод journalctl, который был ротирован на прошлой неделе.
silentwatch-mcp предоставляет эту видимость в виде инструментов MCP, которые ваш ИИ-агент может запрашивать напрямую. Никаких метрических конвейеров, отдельных дашбордов или подписок SaaS.
> claude: which of my cron jobs have silent failures in the last 24 hours?
[MCP tool: find_silent_failures]
3 jobs flagged:
• web-search-refresh — ran 12× successfully but output empty in 8 (66% silent fail rate)
• daily-summary — ran 1× successfully (24× expected); output normal
• audit-snapshot — last success 5 days ago, all subsequent runs returned exit 0 with empty bodyRelated MCP server: task-orchestrator
silentwatch-mcpТри вещи, которые не делают существующие инструменты (Cronitor, Healthchecks.io, Datadog, Prometheus):
Обнаружение тихих сбоев, а не только кодов выхода. Традиционный мониторинг cron предполагает, что exit 0 = успех. Мы проверяем вывод на основе настраиваемых правил: пустой вывод, аномалия длины по сравнению с исторической медианой, ключевые слова ошибки в stdout, несмотря на код 0, аномалия длительности. Задание, которое «успешно выполнилось», но не вернуло ничего полезного — это именно тот тип сбоя, который может скрываться неделями. Мы его находим.
MCP-native, без слоя интеграции. Claude Desktop, Cline, Continue, агенты OpenClaw — любой клиент с поддержкой MCP запрашивает данные напрямую. Никаких плагинов Grafana, API-оберток или ручного разбора JSON.
Мульти-источник «из коробки». Логи OpenClaw в формате JSONL, системный crontab (/etc/crontab + /etc/cron.d/* + crontab -l для пользователя) и таймеры systemd (systemctl list-timers + journalctl) — все четыре бэкенда поставляются в v0.3, поэтому вы можете запустить silentwatch-mcp с любым планировщиком, который у вас есть. Никакой привязки к поставщику.
Создан для SMB-хостеров, запускающих VPS за $40, где Datadog — это излишество, а «бесплатный open-source MCP» — подходящая ценовая категория, хотя обнаружение тихих сбоев столь же ценно и на корпоративной инфраструктуре.
Сервер регистрирует следующие инструменты MCP (полная спецификация в SPEC.md):
Инструмент | Что он делает |
| Перечисляет все известные задания cron со сводкой последнего запуска |
| Подробный статус одного задания: последний запуск, последний успех, процент успешных запусков за период |
| История недавних запусков с таймингами, статусом и фрагментом вывода |
| Задания, которые согласно расписанию должны были запуститься, но не запустились |
| Задания, которые выполнились «успешно», но вывод выглядит подозрительно |
| Недавний вывод логов для одного задания |
Ресурсы:
cron://jobs — список всех заданий (манифест)
cron://job/{id} — манифест отдельного задания + недавние запуски
cron://run/{id} — экземпляр отдельного запуска с полным выводом
Промпты:
diagnose-overdue — шаблон промпта для диагностики просроченного задания
summarize-cron-health — ежедневный дайджест активности cron + аномалии
v0.3 beta — все 4 бэкенда включены + реальное обнаружение просрочек через парсинг расписания cron (croniter). Бэкенды Mock, OpenClaw JSONL, crontab и systemd готовы к промышленному использованию. 74 теста пройдены. v1.0 — это полировка: релиз в PyPI + GitHub Actions CI + отправка в реестр MCP.
pip install silentwatch-mcp # not yet on PyPI; install from source for now:
pip install -e .Добавьте в ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) или %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"silentwatch": {
"command": "python",
"args": ["-m", "silentwatch_mcp"],
"env": {
"SILENTWATCH_BACKEND": "mock"
}
}
}
}Бэкенды (все четыре поставляются начиная с v0.3):
SILENTWATCH_BACKEND=mock — возвращает примеры данных (по умолчанию для разработки)
SILENTWATCH_BACKEND=openclaw-jsonl — парсит нативные файлы логов OpenClaw в формате JSONL (установите SILENTWATCH_OPENCLAW_LOGS в директорию, по умолчанию ~/.openclaw/cron-runs/); самые богатые данные — полная история запусков + обнаружение тихих сбоев
SILENTWATCH_BACKEND=crontab — парсит /etc/crontab + /etc/cron.d/* + пользовательские crontab (crontab -l); время последнего запуска берется из /var/log/syslog или /var/log/cron (установите SILENTWATCH_SYSLOG для переопределения)
SILENTWATCH_BACKEND=systemd — парсит systemctl list-timers --all --output=json + journalctl -u <unit> для истории запусков; извлекает OnCalendar= в поле расписания
Все бэкенды, кроме mock, корректно возвращают пустые результаты на платформах/хостах, где отсутствуют соответствующие инструменты, поэтому конфигурацию можно безопасно оставлять во всех окружениях.
Сервер регистрируется как silentwatch. Проверка:
Покажи мне все мои задания cron и статус их последнего запуска.
Версия | Область | Статус |
v0.1 | Настройка протокола, mock-бэкенд, все 6 инструментов зарегистрированы с заглушками, тесты пройдены | ✅ Завершено |
v0.2 | Реализован бэкенд OpenClaw JSONL (реальный парсинг запусков cron, обработка некорректных строк, обогащение тихими сбоями) | ✅ Завершено (2026-05-02) |
v0.3 | Бэкенды Crontab + systemd; парсинг расписания cron для реального обнаружения просрочек (croniter); 35 новых тестов | ✅ Завершено (2026-05-02) |
v1.0 | Полировка: релиз PyPI, GitHub Actions CI, отправка в реестр MCP (Glama + PulseMCP), уточненная конфигурация правил тихих сбоев | ⏳ Цель Фазы 1 (неделя 3, 18 мая) |
v1.x | Дополнительные бэкенды (планировщик Cowork, фоновые задачи Claude Code, общая JSON-конфигурация), вебхуки для оповещений | ⏳ Фаза 2+ |
silentwatch-mcp поставляется с 4 бэкендами (mock, OpenClaw JSONL, crontab, systemd). Если ваш планировщик — это что-то другое (AWS EventBridge, GCP Cloud Scheduler, Hangfire, Sidekiq, Temporal, Apache Airflow, Prefect, Dagster или кастомный исполнитель заданий), и вы хотите получить такую же видимость обнаружения тихих сбоев через MCP, это задача для Custom MCP Build.
Уровень | Область | Инвестиции | Сроки |
Простой | Адаптер одного бэкенда для существующего планировщика с документированным API (например, GCP Cloud Scheduler) | $8,000–$10,000 | 1–2 недели |
Стандарт | Кастомный бэкенд + кастомные правила тихих сбоев + интеграция с вашим оповещением (PagerDuty, Slack и т.д.) | $15,000–$20,000 | 2–4 недели |
Сложный | Мульти-бэкенд (федеративный cron по регионам/кластерам/арендаторам) + RBAC + интеграция логов аудита + рабочий процесс дежурства | $25,000–$35,000 | 4–8 недель |
Для заказа:
Напишите на admin@pixelette.tech с темой Custom MCP Build inquiry
Включите: описание вашего стека планировщиков в 1 абзаце + какой уровень вы рассматриваете
Ответ в течение 2 рабочих дней с предложением 30-минутного звонка для обсуждения
Этот сервер также является частью AI Production Discipline Framework — методологии, лежащей в основе проводимых мной аудитов ИИ в продакшене.
Если вы используете ИИ в продакшене и хотите, чтобы сторонний специалист оценил готовность, нашел существующие паттерны сбоев и составил план корректирующих действий — это то, для поддержки чего создан данный MCP. Услуги аудита:
Уровень | Область | Инвестиции | Сроки |
Audit Lite | Одна система, топ-5 находок, письменный отчет | $1,500 | 1 неделя |
Audit Standard | Полный аудит, все 14 паттернов, 5 находок Cs, последующее наблюдение 90 дней | $3,000 | 2–3 недели |
Audit + Workshop | Стандартный аудит + 2-дневный командный воркшоп + первый ежемесячный аудит включен | $7,500 | 3–4 недели |
Тот же email: admin@pixelette.tech с темой AI audit inquiry.
PR приветствуются. Структура намеренно плоская, чтобы упростить добавление кастомных бэкендов — см. src/silentwatch_mcp/backends/ для примеров.
Чтобы добавить новый бэкенд:
Создайте подкласс CronBackend в backends/<ваш_бэкенд>.py
Реализуйте list_jobs, get_job_runs, tail_logs
Зарегистрируйте в backends/__init__.py
Добавьте тесты в tests/test_backend_<ваш_бэкенд>.py
Отчеты об ошибках и запросы функций: открывайте issue на GitHub.
MIT — см. LICENSE.
AI Production Discipline Framework — шаблон Notion, $29
SPEC.md — полный дизайн сервера
Model Context Protocol — обзор протокола
Создано Temur Khan — независимым специалистом по ИИ-системам в продакшене. Контакт: admin@pixelette.tech
Returns jobs whose schedule indicates they should have run but haven't, beyond a grace window.
| Name | Required | Description | Default |
|---|---|---|---|
| grace_minutes | No | Tolerance to avoid flagging jobs about to run (default 5) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It discloses the core behavior (returns overdue jobs) but does not mention whether the operation is read-only, performance implications, or pagination. Adequate but not thorough.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single sentence that is efficient and front-loaded with the core purpose, no redundant words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with one optional parameter and no output schema, the description is largely complete. It could clarify what 'schedule indicates' means or the output format, but overall it provides sufficient context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With 100% schema coverage, the description adds value by explaining 'beyond a grace window' which directly connects to the grace_minutes parameter, providing context beyond the schema's technical description.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool returns overdue jobs with a grace window, distinguishing it from siblings like find_silent_failures (different failure mode) and list_jobs (all jobs).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for checking missed jobs but lacks explicit when-to-use or when-not-to-use guidance, nor does it mention alternatives among siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Jobs that returned exit code 0 but output was flagged by silent-fail rules (empty output, length anomaly, error keywords, duration anomaly).
| Name | Required | Description | Default |
|---|---|---|---|
| window_hours | No | Lookback window in hours (default 24) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided; description only lists detection criteria. It does not disclose behavioral traits like read-only nature, prerequisites, or potential side effects.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single sentence efficiently conveys purpose but lists multiple anomaly types in a somewhat dense manner. No wasted words, but readability could improve.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given one optional parameter and no output schema, the description lacks details on return format, pagination, or usage context. Incomplete for a search/filter tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% with clear description for window_hours. The tool description adds no extra meaning beyond the schema, so baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it finds jobs with exit code 0 flagged by silent-fail rules, using specific verb and resource. However, it does not differentiate from siblings like find_overdue_jobs.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance on when to use this tool vs alternatives. The description implies usage for detecting silent failures but lacks when-not or alternative tool references.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent run history for a job (newest first) with timing, exit code, status, silent-fail indicators, output snippet.
| Name | Required | Description | Default |
|---|---|---|---|
| job_id | Yes | Job identifier | |
| limit | No | Max runs to return (default 20, max 500) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so the description carries the full burden. It discloses ordering, data fields, and indicators (e.g., silent-fail), but does not mention side effects, rate limits, access requirements, or return format specifics. This is adequate but not thorough for a tool with no annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single sentence with all key information front-loaded (recent, newest first, data fields). No wasted words; every part adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema exists, so description should explain return structure. It lists fields but does not specify if results are an array, pagination behavior (beyond limit), or error handling. Adequate for a simple list but incomplete for comprehensive understanding.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so baseline is 3. The description does not add meaningful information beyond the schema: 'job_id' and 'limit' are already described in the schema with default and max values. No additional context for parameter usage is provided.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly specifies the resource ('run history for a job'), action ('get'), ordering ('newest first'), and included data fields ('timing, exit code, status, silent-fail indicators, output snippet'). It effectively distinguishes from sibling tools like get_job_status or tail_job_logs.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No explicit guidance on when to use this tool versus alternatives like get_job_status or find_silent_failures. The description implies usage for recent runs but lacks when-not-to-use conditions or comparisons.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Detailed status for one job: last run, last success, success rates over 24h + 7d, overdue state, silent-fail indicators on the last run.
| Name | Required | Description | Default |
|---|---|---|---|
| job_id | Yes | Job identifier from list_jobs |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided; the description describes return fields but does not disclose any behavioral traits such as read-only nature, authentication needs, or cost. For a read-like tool, this is a notable gap.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The entire description is a single concise sentence that front-loads the core purpose and lists specific details without any wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's low complexity (one parameter, no output schema), the description adequately covers return values. It could mention error handling or prerequisites, but is largely complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Input schema has 100% description coverage with 'job_id' documented. The description adds no further parameter information, so a baseline score of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Detailed status for one job' and enumerates specific fields (last run, success rates, overdue state, silent-fail indicators), distinguishing it from siblings like find_overdue_jobs or get_job_runs.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for obtaining a comprehensive snapshot of a single job's health, but does not explicitly state when to use this tool over alternatives or provide exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Enumerate all known cron jobs with last-run summary. Returns id, name, schedule, last run time + status, runs/successes in last 24h, silent-fail count, overdue flag.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Despite lacking annotations, the description transparently lists all returned fields, including last-run summary and overdue flags. This adequately discloses the read-only behavior and output structure.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that efficiently conveys the tool's purpose and output. No extraneous information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's low complexity (no parameters, no output schema), the description covers its functionality and output comprehensively. It could mention it as a read-only operation, but the field list suffices.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With zero parameters, the description adds value by detailing the output fields, compensating for the absence of an output schema. It provides richer semantics than the empty input schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool enumerates all known cron jobs with a last-run summary, specifying the returned fields (id, name, schedule, last run time + status, etc.). It distinguishes itself from sibling tools like find_overdue_jobs and find_silent_failures, which target specific subsets.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
While the description implies general-purpose enumeration, it does not explicitly state when to use this tool versus siblings like get_job_status or get_job_runs. No exclusion criteria or alternative recommendations are provided.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Most recent N log lines for a job (newest last).
| Name | Required | Description | Default |
|---|---|---|---|
| job_id | Yes | ||
| lines | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. Only states the result order and count, but omits read-only hint, error handling, or limitations like max lines.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single sentence, no wasted words. Efficient but could include more contextual info without becoming verbose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Adequate for a simple tool with 2 parameters and no output schema. Lacks details on behavior for edge cases and result format, but core purpose is clear.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%. Description hints at 'lines' parameter ('N log lines') but does not explain 'job_id' or provide format/constraints for either parameter.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states the action ('Most recent N log lines'), resource ('a job'), and ordering ('newest last'). Distinguishes from siblings like get_job_status or list_jobs.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No explicit guidance on when to use this tool vs siblings. Does not mention exclusions or alternatives despite related tools (e.g., find_silent_failures, get_job_runs).
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.3.0find_overdue_jobsfind_silent_failuresget_job_runsget_job_statuslist_jobstail_job_logsScored across 6 tools
Each tool addresses a distinct aspect of job monitoring: overdue detection, silent failure detection, run history, job status, job listing, and log tailing. No overlap in purpose.
All tool names follow a consistent verb_noun pattern with underscores, using clear verbs like find, get, list, and tail. No mixing of conventions.
Six tools cover the core functionality of a job monitoring server: listing, anomaly detection, status, logs. Neither too few nor too many for the scope.
The tool set provides complete coverage for monitoring cron jobs: discovery, anomaly detection (overdue and silent failures), status, history, and logs. No obvious gaps for a read-only monitoring use case.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Control plane for autonomous software labor. Agents claim objectives over MCP with audit trail.
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Uptime, SSL, DNS and domain monitoring you can talk to from Claude or any MCP client.
